Analysis of agricultural technology adoption: the use of improved maize seeds and its determinants in Ethiopia, evidence from Eastern Amhara

Abstract

Agriculture is the major source of income and employment creation for majority of least developed countries. Ethiopia as one of least developed countries of sub-Saharan Africa, majority of its citizen’s livelihood is predominantly dependent on agricultural products, more specifically on the production of cereal crops (Alemayehu et al. in Adv Crop Sci Technol 6(1):1–9, 2018). This study is conducted to identify the determinants of household to adopt and use improved maize seed, one of highly cultivated cereal crops in Eastern Amhara of Ethiopia. The data used in this study are secondary data collected by the agriculture office of Amhara regional state from 3864 household farmers living in the study area in 2022 fiscal year. To address the objective of the study, both descriptive and econometric methods of data analysis are applied. Because of the binary nature of adoption to be either adaptor or non-adaptor, probit regression model is applied. The probit model result shows that being male-headed household, rural credit provision, access to irrigation and the use of fertilizers affect the tendency of adopting improved maize seeds are positively related, while total arable land size owned by farmers negatively affects adoption status. Based on this result, the researcher recommends that government and other institutions should be invited to facilitate and directly involve in providing credit, irrigation, and fertilizers in the study area.

Introduction

Ethiopia is one of the least developing countries in Africa in which majority of its population is dependent on agricultural activities of crop production and animal rearing (Wassihun et al., 2022). Moreover, about 80% of GDP, more than 60% of employment and 70% of export earning is generated from the agricultural sector even if it follows backward method of production with obsolete technologies. So to increase the welfare and living standards of Ethiopian citizens, more emphasis must be given for these sectors and modern technologies like, fertilizers, improved seeds and pesticides has to be used by farmers in rural areas. Ayele (2009) indicates that increasing the productivity of rural farmers can lead to improvements in living standards and welfare of Ethiopian citizens if it is supported by the introduction of highly advanced new agricultural technology adoption either by dissemination of new pesticides and insecticide or by providing new and productive improved seeds.

Maize is one of the most important cereal crops useful to improve food security status of Ethiopian citizens after recurrent droughts and famines in the late 1884 (Abate et al., 2015). Currently, about nine million household farmers are producing cereal of maize and this crop is the first crop to be produced by the highest number of farmers in Ethiopia with the highest productivity rates when compared with other common crops like Teff.

The productivity and maize output at time is estimated to be higher than 3 metric tones/hectare and this is the second highest maize productivity among African countries next to South Africa (Abate et al., 2015). The rural land area covered by improved maize varieties in Ethiopia shows highest growth from time to time with specific example of 14% productivity in 2004 to 40% productivity increments in 2013 (Teferi et al., 2020), and the rate at which the use of fertilizers and improved seeds is advanced from 16 to 34 kg/ ha in the same period stated above. In Ethiopia, the ratio of farmers to extension workers is given by the ratio 476:1 which is very low and some how good when compared with some African countries like Kenya with ratio of 1000:1 (one thousand farmers are supported by only one extension worker), Malawi with one extension worker supporting 1603 rural household farmers and Tanzania with ratio of 2500 farmers/1 extension workers.

So this advancement in the use of improved maize varieties and fertilizers, coupled with increased extension services and the reduction in the occurrence of droughts are the primary forces in promoting the accelerated growth in maize productivity in Ethiopia. Ethiopia took a homegrown solutions approach to the research and development of its maize and other commodities. The trend from Ethiopia’s past production of maize indicates that sustainable improvements in agricultural investments and research development are supporting and positively affecting growth of cereal crop productivity of small holder farmers at micro level and growth of agriculture share to GDD at large (Jayne et al., 2003).

Studying the level of agricultural productivity small holder farmers specifically relating with cereal crops productivity has to be the primary objective of any researchers (Molla, 2017), since study findings in this thematic area can enhance the food security problems of African countries. Based on this idea, many numbers are conducted to study the determinants of productivity of small holder farmers in and some of those studies are presented below.

Household farmers of rural Eastern Amhara regions are mainly dependent on agriculture most importantly practicing crop production and animal husbandry (Degefa, 2020). So the livelyhood of those farmers is highly affected by the agricultural productivity of main crops produced in the area such as maize, teff and sorghum. Even if maize is the major cereal crop in the study area, little attention is given and more is done on the productivity of teff crop. Using improved seeds is (Zeng et al., 2018) the most significant determinant to in scaling up the productivity and yield of household farmers which in return will enhance the food security and welfare of rural households. And hence the use of those improved seeds is highly affected by farmers willingness to pay and many determinants factors are identified by different researchers. Gebre (2020) conducted a study on the farmers willingness to pay of improved sesame seed driller in Humera district and he found that level of education, family size, total agricultural yield and attitude towards technology adoption to be the most significant variable in affecting willingness-to-pay status of households.

Another study by Yirgalem and Biniam conducted to identify the determinants farmers’ willingness to pay for improved teff seed in Northern Ethiopian Yilmana- district (Tilahun & Tadesse, 2022). Based on bi-variate probit model result, education level, gender, credit extension contact and on-farm income, have significant and positive impact on the farmers status of willingness to pay for maize improved seed.

In Sidama Regional state farm households willingness to pay money in cultivating forage seeds (Mulatu, 2016) is investigated to declare that high farm land size and level of households asset opportunities of household heads to participate in training, providing access to credit, inputs like fertilizers and extension services to determine farmers willingness to pay in the study area. Another study by Tesfaye focuses on analyzing willingness to pay of farmers for water lifting technologies in Bahir-Dar zuria district (TESFAYE, 2015) and the findings of the study obtained from multi-nominal logit model shows that, access to product market, agro-ecological zone and land size, household income, quality of soil and sex of the household head, family size, education level, access to credit, off-farm activity are identified to the significant variables that can affect the probability of household farmers relating to willingness to pay for water lifting technology.

DBE catalyze indicates that market prices for raw materials, windshield wiper washers and motor bobs are between the upper and lower limits of farmers' WTP limits, but farmers are willing to pay advanced for pulleys than the existing market indicate. I'm here. Additionally, the mean values of the public interview are lower than those of the double-tested interview (Wassihun et al., 2022).

Another research by Mulatu (2016) conducted on factors that can promote improved forage seed usage on private individuals and enterprises level and the factors affecting them to enjoy the opportunities are identified to include variables like the provision of extension workers technical supports, governments policy to give emphasis for those businesses and provision and access to inputs including land and the running raw materials access to irrigation are statistically significant variables.

So, the main objective of this study is identifying the determinants of agricultural technology adoption through the use of improved maize seed and chemical fertilizers which is important issue to take action on the productivity and welfare improvement process of rural households in Ethiopia. The significance of this study is multidimensional since it will first benefit the farmers by giving recommendation for concerned government officials to fill farmers need after identifying the significant variables power full to affect household farmer’s decision of being adapters and buy or not to adopt improved maize seed. In addition to this, the finding of this study will benefit the urban dwellers by increasing the supply of maize and can access maize product at least cost resulting from the use of improved seeds.

Review of related literature

Theoretically, the introduction of agricultural technology is expected to improve household welfare by increasing productivity of agriculture and poverty reduction (Hawas & Degaga, 2023). However, the potential impact of technology adoption depends on whether farmers adopt the agricultural technology and, if so, on the rate of adoption of the agricultural technology. This is usually measured by the time it takes for a certain percentage of the system's members to adopt an innovation. Furthermore, innovations that individuals perceive as having a high relative advantage, high compatibility, low complexity, separability, and observeability have higher adoption rates (Legesse et al., 2019; Korir et al., 2023).

Adoption decisions are generally assumed to be the result of optimization of expected returns (Massresha et al., 2021), which are a function of land allocation, the production capabilities of the technology, input costs, and output prices (Merga et al., 2023). The contribution of agricultural technology adoption to economic growth in terms of productivity and poverty reduction will only be realized if the adopted technology is widely disseminated and used (Mekonnen, 2017; Merga et al., 2023).

The diffusion of an innovation is the result of a series of individual decisions regarding the use of new technology (Mogess & Ayen, 2023). Decisions are often made by weighing the uncertain benefits of a new invention against the uncertain costs of its introduction (Mujeyi et al., 2021). When farmers decide to adopt a particular technology, they need to weigh the impact of introducing the innovation against its economic, social and technical feasibility (Habtewold & Heshmati, 2023). Farmers then evaluate new technologies in terms of the additional benefits of using them.

Although many studies have found that credit has a positive impact on household happiness (Mutungi et al., 2023), there are still studies that have shown that credit has a negative impact on household happiness (Nonvide, 2024).

Existing literature demonstrates the positive impact of technology adoption on productivity, poverty reduction, and living standard around the world (Shita et al., 2023). Similarly, studies in Ethiopia have shown that improved agricultural technology positively impacts productivity and welfare (Shokati Amghani et al., 2023) and improves food security for smallholder farmers (Sinha & Nag, 2023).

According to TEFERA, Ahmed et al. (2023), the use of improved seeds and chemical fertilizers alone will increase crop productivity in Ethiopia by 7.38% and 6.32% annually, respectively. However, despite increasing penetration rates and its positive impact on production and productivity, the majority of rural households in Ethiopia live in poor living conditions. Nowadays, wheat production accounts for 16% of Ethiopia's total grain production area. Approximately 36% of grain-growing households in Ethiopia are directly dependent on wheat cultivation.

However, the national average productivity of wheat is 1.83 t/ha (Tesfay, 2023) and in 2018 was 2.7 t/ha. Wheat production is also expected to be 2.77 tonnes per hectare, increasing the total cultivated area to 1.66 million hectares. 2019/2020 harvest season. Despite this, Ethiopia was unable to meet its domestic wheat demand. It produces approximately 4.6 million tons each year, but its consumption exceeds its production level (i.e., 6.3 million tons per year) (Tesfaye & Tirivayi, 2018; Tilahun et al., 2023). In addition to low productivity, the demand for wheat is increasing in both rural and urban areas of Ethiopia (Tilaye et al., 2023), resulting in people being unable to meet the growing demand and contributing to the existing poverty situation in the country.

According to Tolola (2023), many countries have contract builders such as Model Builders in Ethiopia and Master or Progressive Builders (Verkaart et al., 2017) in Malawi to assist government advisors in technology transfer and information provision.

According to Wordofa and Sassi (2020) and Wossen et al. (2017), model farmers are farmers who generally have relatively good resources, are early adopters, and are able to provide at least 70% of the technology packages provided by agricultural extension systems. It is defined as a farmer who has introduced the system and meets the following conditions. Their influence in the farming community is recognized by the development actors (grassroots workers). Many views are raised on the impact of credit on household welfare. While savings provide security for future well-being, credit to households is a means of escape from current financial constraints used to maintain or improve well-being (Baig et al., 2023).

The study by Feyisa (2020) was carried out using the Tobit model on the determinants of the implementation and intensity of barley malting technology in Limna Bilbiro, Shashemene and Kofere districts of the Oromia National Regional State (Gelata et al., 2024). The results of the Tobit model show that social participation and overall cultivated farm size have a negative impact on the adoption of barley malting technology, while plot size, profit level, distance from market, and farmers' attitude towards barley malting technology have a significant effect (Assaye et al., 2023).

Another study was conducted by Kebede (2023) on the factors influencing the adoption and intensity of barley malting technology package in Marga woreda, southern Ethiopia using Tobit model. The results of the Tobit model show that factors influencing adoption and its intensity include education, family size, land size, access to credit, co-operative membership, access to training, and access to demonstrations. The total number of livestock units and the distance to the animals were shown to be included in the nearest market. These in one way or another significantly influenced farmers' adoption decisions and adoption intensity (Zegeye et al., 2022) (Fig. 1).

Fig. 1

Conceptual frame on agricultural technology adoption

Methodology of the study

Description of the study area

The Amhara Region (also known as Amhara National Regional State is one of the 13 regions found in Ethiopian federalism institutional setup best known to be the homeland of ethnically Amhara people. The capital of Amhara region is Bahir Dar City in which the Regional Government of Amhara is seated. The region is gifted with man-made and natural places used for practicing tourism including Lake Tana, source of Blue Nile River, Tis-Esat Falls, Gondar palaces, Semien Mountains National Park and creative Rock cut churches of Lalibela. Amhara is a vast area bordered by afar region in the east, Benishangul-Gumuz to the southwest Sudan to the west, Oromia to the south and Tigray to the north.

The study area, Eastern Amhara is a lowland area bordering with Tigray Region in the north and North Shoa zone even though it includes some lowland parts of this zone, in the south and Afar region in the East, including districts of Raya Kobo, Woldia, Habru, Gubalafto, Ambasal, Kalu, Tehuledere and Ataye provinces. Household farmers of this area more practice animal husbandry and crop production with wheat, teff, sorghum and maize as major cereal crops used for household self-consumption and market (Fig. 2).

Fig. 2

Map of the study area

Data type, source and description of variables

The data used in this study are secondary data collected by Agriculture office of Amhara regional state on the household farmers of Eastern Amhara areas. So the data are collected from 3864 household farmers living in the eastern and lowland part of part of the region north and South Wollo provinces including Raya Kobo, Habru, Gubalafto, Antsokia Gemza and Ambasal districts. The time of this data collection was the 2022 Ethiopian fiscal year and all the data relating to household farmers indicate their production and other status of this year. The description of variables used in this study is expressed below (Table 1).

Table 1 Description of variables and expected sign

Sampling procedure

The data used in this study are collected from household farmers living in Eastern Amhara region, more specifically on the three Zones of North Shewa, North Wollo and south Wollo provinces. To gather the data, agriculture office of Amhara regional state uses stratified and systematic random sampling techniques. Farmers in the study area are classified based on their living district and then individual household heads are selected randomly. Using this procedure, about 3864 sampled households are included to be respondents of the data collection process. To obtain a representative sample, a multi-stage probability sampling method. In the first stage, the population had been classified into two strata based on their area of residence: highland and lowland settlers in both zones. In the second stage, systematic random sampling was applied in each stratum to select four lowland areas and four highland areas from each zone. In the third stage, representative kebelles were selected in each stratum proportionally using systematic random sampling. In the final stage, households in the selected kebelles were selected proportionally from each stratum using systematic random sampling.

Methods of data analysis

In this study, both descriptive and econometric methods of analysis with quantitative research design are applied. First, the data obtained from sampled household farmers are analyzed using measures of variation, measures of central tendency, with the help of tables and figures. In addition to this, econometric analysis which can identify the relationship between agricultural technology adoption and its determinants is analyzed with the help of econometric model of probit regression.

Model specification

Dependent variable of this study, agricultural technology adoption is binary by its nature. Because of this categorical nature of the dependent variable, the non-linear binary logistic or probit regression models are appropriate for this case. The only difference between logit and probit model (Zegeye et al., 2022) is the type of CDF they follow, since logit model has a logistic cumulative distribution function while probit model follows normal cumulative distribution function. In this research, it is assumed to have normally distributed data and probit model is selected to be appropriate model. After identifying the most important variables that can determine farmers willingness to pay for improved maize seed, the probit model specified as follows.

The probit regression model is the standard normal cumulative distribution function (CDF) expressed as an integral and this model is preferred to other binary choice model by assuming the data will follow normal CDF and even robusted regression can be applied to capture the effect on non-normality in estimation procedure (Wudu et al., 2021):

$$G(z) = \Phi (z) = \int_{ - \infty }^z {\phi (z)} {\text{d}}z,$$

$$\phi (z) = (2\pi )^{ - 1/2} e^{ - z^2 /2}$$

where $= \frac{1}{{\sqrt {2\pi } }}e^{ - \frac{1}{2}z^2 } .$

Then, G(z) becomes:

$$G(z) = \Phi (z) = \int_{ - \infty }^z {\frac{1}{{\sqrt {2\pi } }}e^{ - \frac{1}{2}z^2 } } {\text{d}}z,$$

where Z = XB + e.

In this study, the probit model will be used to identify the determinants of household adoption status for improved maize seed in Eastern Amhara. To specify the relationship between the dependent variable and the independent variables, the study considers households agricultural technology adoption status as dependent variable and to take only two forms of either adapters or non-adapters:

$${Y}^{*}=\beta \text{X}+\varepsilon ,$$

where $\beta$ a 1 × m is vector of returns to characteristics and $\varepsilon$ is stochastic error term with zero mean. With representative sample of the population, X can be used to predict adoption status. The dependent variable is household’s adoption status represented in the model as adapters (Y = 0) and not adapters (Y = 1) if they do not want to use improved seeds of maize.

Consider an econometrics model:

$${Y}^{*}={\beta }_{0+}\sum_{j=1}^{n}{\beta }_{j}{X}_{ij}{+ \varepsilon }_{i},$$

where ${Y}^{*}$ = is a "latent" variable as a proxy variable for adoption status which is the dependent variable of this study:

$${{Y}}_{{i}}=\left\{\begin{array}{ll}1&\quad {\rm if}\, {{Y}}^{*}>0\\ 0,&\quad {\rm Otherwise}.\end{array}\right.$$

${Y}^{*}$ is the probability of a person being adapters,

$${P}_{i}={\rm prob}\left({Y}_{i=1 }\right){\rm prob}\left[{\varepsilon }_{i}>-({\beta }_{0}+\sum_{j=1}^{n}{\beta }_{j}{X}_{ij})\right]$$

$$= 1-\text{F}\left[{-(\beta }_{0 }+\sum_{j=1}^{n}{\beta }_{j}{X}_{ij})\right],$$

where F is the cumulative distribution function to represent $\varepsilon$. If the distribution of $\varepsilon$ is symmetric, since 1 − F (− Z) = F(Z), we can write as follows: ${P}_{i}$ = F${(\beta }_{0 }+\sum_{j=1}^{n}{\beta }_{j}{X}_{ij}$).

$$\begin{aligned} {\text{ATA}} & = \beta 0 + \beta 1\,{\text{land}}\,{\text{size}} + \beta 2\,{\text{fertilizers}} + \beta 3\,{\text{Gend}} + \beta 4\,{\text{Irrigation}} + \beta 5\,{\text{education}} + \beta 6\,{\text{credit}} + \varepsilon i \\ & = XB + e, \\ \end{aligned}$$

where APA = agricultural technology adoption = 0 if adapters

1, otherwise,

Land size = households total maize production measured in quintals,

Fertilizers = gross agricultural production of cereals other than maize measured in quintals,

Gender = gender of household heads = 1 if male headed.

0 otherwise,

Irrigation = dummy variable for households access to irrigation = 1 if yes.

0 otherwise,

Education = the household heads year of education to be considered as a continuous variable,

Credit = dummy variable for access to credit = 1 if accesses credit.

0 otherwise,

Ε = The error/disturbance term of the model.

Results and discussion

Descriptive analysis

Descriptive summary

The descriptive summary of both the categorical and continuous variables are analyzed using measures of mean, minimum, maximum, standard deviation, frequency, percentage, skewness and kurtosis measures as displayed in the two tables presented below.

The tabular presentation of categorical variables (Table 2) can indicate the number of observations in each group of the dummy variable. In this regard, out of the total of 3864 household farmers, 1982 (51.29%) are adopting and using improved maize seed and the remaining 1882 (48.71%) are not adopting and using this improved seed because of their own personal factors identified in the probit regression model presented in the econometric analysis part (Table 9). Out of the total sample, only 842 (21.8%) are female headed, 1212 (31.4%) had access to irrigation for crop production specially maize, 1654 (42.8%) had access to credit from banks and micro finance institutions. We can observe that only small proportion of households are accessing irrigation and credit to improve their livelyhoods and welfare, thereby needs high involvement by governmental and nongovernmental institutions. More specifically, the agricultural office of Amhara region is needed to directly involve in providing multi-directional supports to enable household farmers to use more advanced technologies that can expand access of water irrigation and this finding is supported by Molla (2017) and Mengistie and Kidane (2016).

Table 2 Summary statistics of dummy variables

In addition to categorical variables included in the model, the descriptive statics of continuous explanatory variables are presented in the above table (Table 3.). Based on this result, the land size has respective minimum and maximum values of 20, 10, 61 and 95. The mean of fertilizers used in production of maize is 36.13 while its average value is given to be 52.69. Education level of households’ ranges from 0 to 14 indicating that the farmers’ education level is has a maximum variance starting from illiterate persons to fourteen years of education to mean that individuals who completed their secondary and preparatory schools are practicing traditional farming agriculture. The skewness and kurtosis values need more explanation since those values are helpful to understand the distribution of observations to be either normal or not. A negative skewness value indicates that variables are skewed to the left and positive values tells the present of positively skewed distribution.

Table 3 Descriptive summary of continuous variables

The kurtosis value of for education is more than three (leptokurtic distribution) to mean that few observations lied in the middle of distribution (around the mean) and more observation has an extreme education status of very high level and low status or being illiteracy level. The remaining variables age, family size and dependency ratio has a kurtosis less than three (platikurtic distribution) tell us (Tadesse, 2019), the distribution of less of those observations lies on the outliers than the normal distribution.

Cross-tabulation of adoption status with determinants

Adoption status with access to credit

Theoretically, access to credit is expected to reduce poverty by adopting mode advanced agricultural technologies (Tadelech & Mesfin, 2020) which in return helps households to engage in different productive activities, to generate better income and smooth consumption Gebru et al. (2021). Thus, it is expected that households who have access and make use of credit are more likely to be adapters (Table 4). Therefore, to explain how improved maize seed adoption is determined on the basis of credit access of the head, the access to credit is divided in to two categories. When we look at the composition of the household heads access to credit, 1654 (42.81%) accessed credit and the rest 2210 (57.19%) household heads had not accessed credit in the study area. Out of a total of 1654 household farmers that accessed credit, 1519 (91.82%) are adopting and using improved maize seeds, and only 135 (8.18%) households farmers are not adopting agricultural technology even if they are accessing credit. To see the deep-rooted impact of credit provision of agricultural technology adoption, it is better to consider the figure of farmers which had which had not accessed credit. Out of the 2210 household farmers who had no credit access, only 463 (about 20%) are adopting and using agricultural technologies and the remaining 80% are not adopting improved maize seed because of lack of credit access.

Table 4 Agricultural technology adoption and access to credit

Adoption status with access to irrigation

Therefore, as presented above (Table 5) comparing the adoption status with access to irrigation, out of a total of 1212 respondents who accessed irrigation, about 60% (728 farmers) household farmers are adopting for improved maize seed and the remaining 484 (40%) household farmers are willing to pay and non-adapters of improved seed of maize among the samples of respondents taken from Eastern Amhara. This indicates that access to irrigation increases the probability of being adaptor for improved seeds. This is because improved maize seeds are more productive when supported with irrigation because of the water absorbing capacity of improved seeds as in the findings of researches conducted by the findings of Emuru (2015) and Gebre (2020).

Table 5 Access to irrigation and adoption status.

Adoption status and education level

Education affects adoption status of the household in different ways. The literature shows that education increases the human capital stock, which in turn increases labor productivity and the motive to use technological methods of production like improved seeds and fertilizer (Mengistie & Kidane, 2016). In this study (as presented in Table 6), increasing educational level increases the farmers adoption status of improved maize seed and the average of years of study for the adaptor and non-adaptor is 8.72 and 2.90, respectively, showing significant difference.

Table 6 Household head education level and adoption status.

Econometric analysis

Diagnostic tests

Testing data for multi-collinearity

If there is perfect collinearity among the determinants of improved maize seed adoption, our estimation may face inaccuracy from its estimation results and hypothesis testings. This means if there is multi-collinearity problem, the coefficients and marginal effect results obtained from the probit model are inaccurate and can have a meaningful interpretation. In addition, the hypothesis testing result to identify statistically significant variables will also face accuracy problem and we cannot know whether variables are powerful to adoption status. So we have to check and the model has better to be free from violation of multi-collinearity. In this study, VIF is used to check the presence of multi-collinearity (with null hypothesis Ho: there is no multi-collinearity and decision rule of failing to reject Ho if mean VIF is less than 10).

As we can see from the above table (Table 7), the mean VIF is given to be 1.01 which is much less than 10 and hence we cannot reject the null hypothesis of no multi-collinearity and hence there is no multi-collinearity problem in this study, thereby meaningful estimation results and statistical significance tests are expected to there within the regression model.

Table 7 VIF as a test for multi-collinearity. Mean VIF 1.01

Heteroskedasticity

If there is no constant variance among different segments of a population which are targets of the study, we can say that there is heteroskedasticity problem to declare the absence of constant variance of the error term. This is a deviation from the assumptions of Gauss–Markov theorem which states that constant and minimum variance of error term must be obtained to have meaningful estimation result. Again we have to check either there is constant variance or not using Breusch–Pagan/Cook–Weisberg test for heteroskedasticity.

Ho: Constant variance

Chi2(1)            = 82.33

Prob > Chi2      = 0.0000

The null hypothesis of this test is Ho: there is a constant variance to be rejected if the probability value is less than 5% 0r 0.05. So in our study, the p value of 0.0000 helps us to reject the null hypothesis and this indicates the presence of heteroskedasticity problem. As a solution to this violation, we applied the robusted probit regression model which can provide accurate estimation and hypothesis testing result by capturing the effect of heteroskedastic variance of error term.

Normality test

Shapiro–Wilk W test for normal data

Variable |            Obs          W                   V                 z                  Prob > z

r |                        3864         0.80385          519.813        16.396         0.00000

The normality test is considered with Shapiro–Wilk test having the null hypothesis of Ho: the error term is normally distributed. The probability of less than 5% (which is 0.0000) indicates that the error term is not normally distributed. This is because of the categorical nature of the dependent variable in the probit model and hence the error term follows Bernoulli distribution. To capture the effect of non-normality of the error term, robusted regression is applied in this study.

Estimation result from the probit model

The first task of researchers in this econometric model specification is identifying which variables are statistically significant and which are not. So the Z value and probability values are a criteria to select important and statistically significant independent variables and hence absolute value of Z value greater than two when (|Z| >= 2), H0:β_j = 0 is rejected and X_j is statistically significant as stated by the rule of thumb) and probability value less than five percent (p < 0.05) are criteria to have significant independent variables. Using this criterion, all variables except education are statistically significant to affect adoption status of household farmers.

As we can see from the above table (Table 8), the logistic regression estimation result tells us adoption status for improved seed of maize is positively related with access to credit, use of irrigation and use of fertilizers while it is negatively related with land size and being female-headed households. This means the probability that household farmer are who accessed credit and irrigation are more likely to adopt and use improved maize seed (Mosisa et al., 2011; Alemayehu et al., 2018; Amare et al., 2016).

Table 8 Estimation result of probit model

In opposition to this, accessing more arable land is negatively correlated with less adoption status of farmer. This means when farmers have more arable land, they are not sensitive to adopt agricultural technology since natural production by itself sufficient when compared with farmers who have small land size. This finding is supported by Chekole and Ahmed (2023). The results given in the form of coefficient of probit model can indicate only the relationship existed between willingness to pay and other co-variates with the sigh of coefficients, but cannot interpret the results in terms of magnitude to measure the unit change in probability of willingness to pay and marginal effect results are so important to measure the change in probability of success when the independent variables are changed by one unit if it is continuous and the difference in intercepts of determinants if it takes binary form.

The marginal effect results of probit model

The marginal effect results of the probit model now can be interpreted as a unit change in probability of adoption status when the independent variable is changed by one unit for continuous variables and the difference in adoption status for attributes of the each dummy or categorical variable (Workayehu & Wortmann, 2011). Starting from continuous variables (Table 9), when the amount of arable land used in production is increased by one hectare, the probability of farmers adopting and using improved maize seed is decreased by 0.004. This means when farmers earn more arable land they are more likely to use traditional maize production system and uses less of improved seed. This result similar to the findings of research done by Teferi et al. (2020) and Tilahun and Tadesse (2022).

Table 9 Marginal effect estimation of the probit model

The other significant dummy variable is access to irrigation in agricultural production. In this regard, the adopting probability of farmers who accessed irrigation (Gebre, 2020, Wassihun et al., 2022) is higher than that of farmers who did not access irrigation by 0.11 probabilities. This means farmers with access to irrigation are more likely to adopt and use the improved maize seeds. This result agrees with the findings of the study of Amare et al. (2016), Mulatu (2016), Gebre (2020) and Chekole and Ahmed (2023).

The last significant variable to affect adoption status of improved maize seed is use of fertilizers and one kilogram rise in fertilizer used by farmer’s is associated with increasing probability of being adapters by 0.001 and who are using fertilizers are more adopters to use improved maize seed. This result is similar to the findings of Worku et al. (2012) and Workayehu and Wortmann (2011).

ROC curve after probit regression model

A receiver operating characteristic curve commonly known as ROC is a graphical illustration that depicted the ability of a binary classifier system to diagnostically test the discrimination among different thresholds either to be varied or not. It was first introduced and developed for operators of radar receivers of military science in 1941 (as presented in Table 10). The prediction power of this method at its best would yield a graph in which the upper left corner or coordinate (0, 1) of the ROC space, representing (no false negatives or 100% sensitivity) and 100% specificity of no false positives. There is perfect classification when the (0, 1) is achieved. A diagonal line (line of no-discrimination) is obtained when random guess gives us the appropriate positive and negative base rates (Fig. 3).

Table 10 Roc test of the probit model

Fig. 3

The ROC curve estimation

The ROC curve of the probit model is given by a bending line to indicate the probability of committing type I and type II is very minimum when type I error is to mean rejecting the true null hypothesis mistakenly to mean in our case refusing to use to use logistic regression and type II error means accepting the null hypothesis which has to be rejected in real and contextualizing in to out study to mean that the probit model was inappropriate to apply for our case but misleadingly prefer to apply in the study. So there is no either type I or type II error and hence the selected probit model is appropriate and no problem in model selection.

Conclusion

This study tried to investigate the determinants of agricultural technology adoption, specifically adoption of improved maize seeds of rural farmers of Eastern Amhara provinces of Ethiopia. With the primary objective of identifying the determinants of adoption status of improved maize seeds, the study used secondary data collected by the agricultural office of Amhara regional state in the year of 2022. After applying both descriptive and econometric analysis of probit model, access to irrigation, credit, fertilizers land size and gender are identified to be the most significant variables to affect adoption status of farmers for improved maize seeds while education level of farmers is statistically insignificant to affect adoption status because of highly informed farmers about the advantage and disadvantage of agricultural technology adoption in the study area. In the study area, being female-headed households is correlated with less probability of adopting new agricultural technologies. Based on this finding, the researchers recommend that the concerned bodies like government offices and non-governmental organizations have to play a significant role in advancing the irrigation access to farmers and increasing the availability of fertilizers. More emphasis is also required to interfere in the community through gender issue interference to improve the adoption status of women and international NGOs are more expected for this interference based on their programs.

Appendix: Do file results of Stata software

The regression result of probit model from STATA software