What matters in adoption of small-scale rain water harvesting technologies at household level? Evidence from Charco-dam users in Nzega, Tanzania

Abstract

Any effort to improve irrigation water availability has an advantage in the crop production processes. Small-scale rainwater harvesting technologies like Charco-dams (Malambo) in Swahili (the language used by most the Tanzanians) are among the interventions proven to overcome agricultural water shortage. Despite its importance in overcoming water stress in most arid and semi-arid areas of Tanzania, some farmers in the areas are still reluctant to adopt such technology. In this regard, this analyzed and discussed the factors influencing the adoption of the Charco-dam rainwater harvesting technology by smallholder vegetable producers in Nzega district, Tanzania. A survey was used to collect the required information from 528 smallholder vegetable producers in the district. A structured questionnaire was employed to collect data from 220 Charco-dam adopters and 308 non-adopters who were used for the analysis. The data collected was substantiated by six focus group discussions (FGDs); one at council headquarters and five at the village level. The Probit model with instrumental variables was used to identify and analyze factors influencing adoption. The study revealed that socioeconomic, farm level and information sharing factors are all important to improve the adoption of Charco-dam technology. The results suggest that any strategy, innovation or policy aimed at increasing the adoption rate of small-scale rainwater harvesting technologies should be designed or formulated by considering socioeconomic, farm-level, environmental, and information sharing aspects.

Keywords:

• small-scale water-harvesting technologies
• Charco-dam
• adoption
• instrumental variable
• Nzega-Tanzania

1. Background

1.1. Introduction

Agriculture and specifically crop production is the most water consuming sub-sector across the globe, simply because water is an essential and primary factor that influences crop productivity and hence guarantees high yield which in return determines small-holder farmers’ wellbeing (Kelemewerk Mekuria et al., 2020; Smith et al., 2011). Apparently, agriculture is inherently sensitive to weather and climate condition, and is among the most vulnerable sectors to the risks and impacts of global climate change and variability (Chartzoulakis & Bertaki, 2015; Kannan & Anandhi, 2020; Mahoo et al., 2015; Rosa et al., 2016; Tzanakakis et al., 2020). It has been agreed that the world is facing a threat from global warming which is causing droughts, flooding, storm damage, long-term shortages of water, worsening soil conditions, desertification, as well as disease and pest outbreaks on crops and livestock (Iizumi & Ramankutty, 2015; Lewis et al., 2018; Rosenzweig et al., 2014, 2002). Consequently, vulnerable areas such as arid and semi-arid areas are experiencing severe losses in agricultural productivity, primarily due to reductions in crop yields (IPCC, 2007; Kang et al., 2009; Ochieng et al., 2016). Therefore, any efforts to improve irrigation water availability have advantages in the crop production processes. According to Gebre and Rahut (2021), East Africa countries experience much higher prevalence of food insecurity due to change in climate and its variabilities as compared to other places in the world. For so many years, small-holder farmers in Tanzania have been facing agricultural water shortage, that directly affecting farm productivity (Jackson et al., 2018). Small-scale rainwater harvesting technologies like Charco-dams (Malambo) in Swahili (the language used by most of Tanzanians) are among the interventions proved to overcome agricultural (for both crops and animals) water shortage. Such technologies observed to have advantages in terms of simplicity in its architectural design, management and control, hence expected to be highly adopted among the arid and semi-arid areas (Awulachew et al., 2005; Mati, 2012; Nissen-Petersen, 2006; Stephens, 2010; Wagner, 2005).

1.2. Statement of the problem

Charco-dam has been used by farmers in several arid and semi-arid areas for domestic use, irrigating the crops as well as for watering the livestock. Hence the use of the water from these dams are differentiated from one area to another based on the water-use, management level, size as well as shapes of the facility, this is mainly due to socio-economic status of the people in particular area (Barron et al., 2009; Hatibu et al., 2000). Unlike other regions in Tanzania (i.e. Shinyanga, Dodoma, Arusha, Tabora, Singida and Mwanza regions), where Charco-dams are owned and managed by community or group of people, Charco-dam in Nzega are owned and managed mostly by individuals mainly for crop farming. Due to the challenges brought by the communal managed and controlled water sources (such as public ponds &dams, wet-valleys & catchments), which include limited frequency and amount of water accessed, and to some catchments limitation on the type of crop to cultivate, small holder vegetable farmers in Nzega are expected to highly adopt this technology, so as to enjoy the unique characteristic of this type of dam to overcome water management and control issues, simply because the production decision is purely based on one decision maker (owner of the irrigation facility) and not majority in the community, and hence enables the adopters to increase yields. However, the assumption is not holding true in Nzega District, whereby adoption of such a technology has been very low (30%-40% across the surveyed wards). Therefore this study analyzed and discussed the socio-economic, farm level and information sharing related factors that influencing the adoption of Charco-dam among small holder vegetable farmers in Nzega district.

1.3. Adoption of technology

Adoption can mean different ways depending on the context of its usefulness. It can be choosing, or continued use of recommended idea or practice, not necessarily be new but novel (Baumüller, 2012; Dasgupta, 1989; Hall & Khan, 2002; Rogers, 2003). According to Rogers (1962), adoption is the mental process that starts when an individual first hears or subjected to the information about certain technology and ends to its final adoption or rejection. Rogers (1983) has distinguished the process in five phases: (1) The knowledge phase: the time when an individual becomes aware of a technology after being exposed to it and gains some idea of how it works; (2) the persuasion phase: this is when the individual starts to interpret the technology (favourable or unfavourable); (3) the decision phase: in which person engages in actions that lead to a choice (adopt or reject the technology); (4) the implementation phase: in this stage the individual puts a technology into use; (5) the confirmation phase: this is an evaluation stage where the individual measure the success or failure of the decision made. In all the five stages, flow of information and particularly how individual perceive the technology has been an important factor for adoption outcomes.

1.4. Measuring adoption of agricultural technologies

Agricultural dynamism due to change in natural conditions, resource availability as well as other socio-cultural issues made adoption of agricultural technologies unavoidable(Cimmty, 1993). Since the introduction of adoption models in agriculture by work of Griliches (1957), agricultural economists have extensively been using the concept in various studies to analyze adoption of various agricultural technologies and their effects on agricultural performance (Adesina & Baidu-Forson, 1995; Akinola et al., 2010; Kimani et al., 2015; Letaa et al., 2014; Nkegbe et al., 2011). There are many models for measuring processes in which an individual engages in adopting a new innovation (Straub, 2009). Since the adoption is an individual decision-making process, then it is the economic choice decision whereby an individual decides to opt for a certain technology if the chosen technology will provide desired expectations, given socio-economic background of the individual. Knowing that, the stages distinguished by Rogers in 2003 are further manipulated to give the two main paradigms; (1) analysis of the adoption of technology as discrete case, and (2) analysis of adoption as a continuous case. In the first case, dichotomous choice models such as logit and probit models are usually employed to identify and evaluate the determinants of adoption for particular agricultural technologies (Aneani et al., 2012; Ayuya et al., 2012; Kaliba et al., 2000; Kijima et al., n.d.; Letaa et al., 2014). While for the second case, different types of censoring and truncated models such as Tobit and Heckman models are usually employed to explore the extent or effects or impact of the adopted technologies on various aspects in farmers’ livelihood and farming performance (Arslan et al., 2013; Baidu-Forson, 1999; Bokusheva et al., 2012; Hailu et al., 2014; Wang et al., 2012).

2. Methodology

2.1. Study area, research design, sampling and data collection methods

A survey design was used to gather the required information from small-holder vegetable farmers in Nzega District. The district is one of the seven districts of Tabora Region in Central Tanzania, receives annual rainfall between 650 mm and 850 mm, with the annual temperature ranging between 28 to 30 °C,¹ while October and July are the warmest and coolest months, respectively. Such rainfall pattern and temperature extremes make the district one of the hottest and driest districts in the region; which is distinguished as a tropical savanna climate and typically pronounced by a dry season, or mostly referred to as semi-arid.

The survey involved 528 small-holder vegetable farmers, 1 district, 5 wards and 5 villages were purposefully selected. The criteria for selection were based on the presence of CDT and the nature and type of crops cultivated (vegetables for this case). Through respective ward and village extension officers, small-holder vegetable producers in each village were identified, and the lists comprised both adopters and non-adopters of the CDT were developed in each village, formula of Yamane (1967) was used to determine required number of farmers in each village (Table 1). Lastly, random selection was done to obtain the number of respondents selected from each village, to get 220 adopters of Charco-dam and 308 non-adopters. Respondents were interviewed using a structured questionnaire, and a total of six focused group discussions (FGDs) were conducted; one at council headquarters and five at village level, to validate and substantiate the information collected during survey.

Table 1. Population and sample selected

Name of Village	Total Vegetable Producers	Sampled Vegetable Producers	Respondents Used for Analysis
			CDT Adopters	CDT Non- adopters
Itunda	102	81	30	46
Ikindwa	138	103	39	62
Shila	152	110	47	63
Busondo	173	121	44	76
Iyombo	184	126	60	61
Total	749	541	220	308

2.2. Theoretical framework

Adoption of agricultural technology involves a number of characteristics that influence an individual decision option, either to adopt or not to adopt the technology (Adesina & Baidu-Forson, 1995; Misaki et al., 2016; Pierpaoli et al., 2013; Rodríguez-Entrena & Arriaza, 2013; Simtowe et al., 2012; Wu et al., 2010). It is noted that for an individual farmer to adopt a certain agricultural related technology, the utility derived from adopting the technology should be higher than the expected utility of not adopting (Afolami et al., 2015). Regarding that, small-holder vegetable farmers in Nzega district are assumed to be rational producers, in the sense that their production choices are in accordance with their preferences, then the equation to model their adoption process is based on utility maximization theory (Adesina & Zinnah, 1993; Sidibé, 2005; Zepeda, 1994). The model assumes that the decision to adopt a particular farm level production technology is based on the maximization of an underlying utility function and a farmer selects his/her production technologies based on his/her expected utility. Therefore a function used to estimate the adoption of Charco-Dam Technology (CDT) can be expressed as:

(1) $C_i^{\ast }=X_i^{\prime }\beta +{\mu }_i$(1)

Where $C_i^{\ast }$ is a latent variable denoting the difference between the utility from adopting Charco-Dam technology ($U_{iA})$ and the utility from not adopting the technology $\left(U_{iNA}\right)$. The farmer will adopt the technology if $\hspace{0.17em}{C}_i^{\ast }=U_{iA}-U_{iNA}>0$. The term $X_i^{\prime }\beta$ provides an estimate of the difference in utility from adopting the technology $\hspace{0.17em}\left(U_{iA}-U_{iNA}\right),$ using the household socio-economic, farm-level and information-sharing fators ($X_i)$ as explanatory variables–which assumed to be exogenous, and ${\mu }_i$ is an error term. Basically under this situation, several choice models such as generalized probability models like Linear Probability Model (LPM), Control Function (CF), or Maximum Likelihood (ML) approaches are appropriate to estimate the adoption equation (1(1) $C_i^{\ast }=X_i^{\prime }\beta +{\mu }_i$(1) ; Afolami et al., 2015; Ding et al., 2011; Ojo, 2004).

However, these methods have some drawbacks; for instance, LPM ignores the effect of binary outcome which may easily provide unacceptable predictions resulting to have over or under estimation (Valente et al., 2018); for CF, it is mostly observed to fit continuously distributed endogenous variables and observed to be weak in estimating discrete endogenous variables (Bontemps & Nauges, 2017; Kalisa et al., 2016); while for ML, despite its efficiency in handling discrete endogenous covariates, it requires a complete parametric specification of dependency of each endogenous covariate on error term, failure to have a proper specification may lead to endogeneity problem (Chesher et al., 2013; Greenland, 2000).

2.3. Analyzing adoption of Charco-dam technology

In cross-sectional data, endogeneity problem can arise when there is measurement error or omission of one or more important variables in the model (Koladjo et al., 2018; Lu et al., 2018). One of the solution to the problem is to use instrumental variable (IV) that replace the endogenous variable with a predicted value that has only exogenous information (Becker, 2016; Chesher et al., 2013). Following the works of Heckman et al. (2006), Bollen (2012), and Angrist and Krueger (2001), the adoption equation (1)(1) $C_i^{\ast }=X_i^{\prime }\beta +{\mu }_i$(1) becomes:

(2) $C_i^{\ast }=I\left(X_i^{\prime }\beta +{\mu }_i\ge 0\right)$(2)

Where $C_i^{\ast }$ is the binary decision variable (to-adopt or not-to adopt CDT), $X_i^{\prime }$ is the vector of explanatory variables (assumed to include at least one endogenous variable), $\beta$ are the parameters to be estimated and ${\mu }_i$ is the error term, that assumed to have zero mean distribution. I(•) is the indicator function taking the value 1 if the latent variable $X_i^{\prime }+{\mu }_i\ge 0$ is to adopt CDT and zero otherwise. To overcome the possibility of mis-specify the endogenous variables in vector $X^{\prime }$, an instrument variable Z was introduced in equation (3)(3) $C_i^{\ast }=I\left(X_i^{\prime }\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\right)$(3) , which assumed to be uncorrelated with the random shock ${\mu }_i$, as; $Cov\left(Z^{\prime },\mu \right)=0$, but correlated with endogenous variable in $X^{\prime }$, i.e. $Cov\left(Z^{\prime },\mu \right)\ne 0$

(3) $C_i^{\ast }=I\left(X_i^{\prime }\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\right)$(3)

From equation (3)(3) $C_i^{\ast }=I\left(X_i^{\prime }\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\right)$(3) , the structural binary-choice model is as follows:

$C_i^{*}=1ifXi\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}0otherwise(4)$

Therefore, this study analysed the adoption of CDT using instrumental variable (IV) equation (4)$C_i^{*}=1ifXi\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}\text{}0otherwise(4)$ whereby $Z^{\prime }$ is the instrumental variable and $\gamma$ is the parameter to be estimated for instrumental variable, the rest of variables are as in equation (3)(3) $C_i^{\ast }=I\left(X_i^{\prime }\beta +Z^{\prime }\gamma +{\mu }_i\ge 0\right)$(3) . For comparison purposes, results from LPM as well as normal probit regressions were also reported (Appendix A). Note, number of labour involved in vegetable production are assumed to be influenced by household-size (for family labour dependants) and the income of household income (for hired labour). Number of labours used in vegetable production assumed to influence adoption of CDT, the number of labours is as well assumed to be influenced by household income (for hired labour) and household-size (for family labour), but the variables household income and household-size are also expected to influence CDT. In such situation, at least one of the explanatory variables can be endogenous, and hence there is a chance of mis-measurement error if otherwise. One of the solutions to overcome this is the use of instrumental variable approach (Lewbel, 2000). Instrumental variables for this case are variables “number of labour used for farm-activities (NoLabourFA)” and “household-income (Hhinc)”, are replacing variable household-size (Hhsize), which is assumed to cause endogeneity problem in the discrete choice model. The description of variables used in the adoption model is given in Table 2.

Table 2. Description of variables used in adoption model

Variables		Description
Dependent Variable
Usecdt (Adoption Model)		1 if respondent used charco-dam technology to irrigate his/her crops; 0 otherwise
Independent Variables
Malehhead	${\alpha }_1$	Male household head, 1 if respondent is a male; 0 otherwise
Agehead	${\alpha }_2$	Age of household head in years
Eduhead	${\alpha }_3$	Household head’s formal education, 1 if respondent attended formal school; 0 if otherwise
Hhsize	${\alpha }_4$	Number of people residing in the household
Landown	${\alpha }_5$	Land ownership; 1 if farm-land is owned by the respondent; 0 otherwise
MemberFin	${\alpha }_6$	Household head’s membership to micro-finance scheme; 1 if yes; 0 otherwise
UseOSW	${\alpha }_7$	Use of other sources of water to irrigate vegetables; 1 if the vegetable farmer used other RW source of water to irrigate his/her crops; 0 otherwise
Lnlandsize	${\alpha }_8$	Log of land-size used to cultivate vegetables in acres
SourceLabour	${\alpha }_9$	Household’s source of labour for the charco-dam related activities, 1 if family I only; 0 if otherwise
Tomato	${\alpha }_{1}0$	Main vegetable crop cultivated, 1 = if cultivated tomato, 0 if not
Cabbage	${\alpha }_{1}1$	Main vegetable crop cultivated, 1 = if cultivated cabbage, 0 if not
Sweetpepper	${\alpha }_{1}2$	Main vegetable crop cultivated, 1 = if cultivated sweet-pepper, 0 if not
ScarletEggplant	${\alpha }_{1}3$	Main vegetable crop cultivated, 1 = if cultivated scarlet-eggplant, 0 if not
LeafVege	${\alpha }_{1}4$	Main vegetable crop cultivated, 1 = if cultivated leaf-vegetables, 0 if not
MemberDeve	${\alpha }_{1}5$	Household head’s membership to community development groups in the village/ ward; 1 if yes; 0 if no
Extser	${\alpha }_{1}6$	Use of extension services, 1 if often; 0 if rarely or not at all
Radio_listen	${\alpha }_{1}7$	Household head’s radio listening behaviour; 1 if often; 0 if rarely or not all

Source: Research Data, 2018

3. Empirical results

3.1. Characteristics of adopters and non-adopters of CDT

The summary statistics of the socio-economic, farm-level and information-sharing factors for adopters and non-adopters of charco-dam technology (CDT) are given in Table 3. It was observed that, 14 out of 22 (64%) of the compared variables were statistically insignificant, this gives an indication that the two groups are likely homogenous and therefore can statistically be compared. However, for the socio-economic variables, it was observed that male household heads occupied a larger share in both overall of the surveyed samples (86.93%), as well as for adopters (90%) and non-adopters (84.74%) categories. The difference of about 5% between the two groups was significant at 10%. No statistical difference observed for the households with formal education between the two groups (adopters and non-adopters of CDT). Mean age for non-adopters of CDT observed to be younger (37.43 years) than that of adopters (39.51%) and the difference is significant at 5% level. Mean household size for CDT-adopters observed to be higher (6.70) than that of non-adopters (3.98), and the difference were significant (at 1%). Further, results show that, the mean annual off-farm incomes for CDT-adopters was 1,361,548/ = TZS) while that of non-adopters was 1,172,404/ = TZS, these results were statistically insignificant. For farm-level variables, no statistical difference observed between adopters and non-adopters for the individuals who owns the vegetable cropping land. It is also observed that, 56.36% of CDT adopters are involved in various micro-credit schemes while only 26.30% of non-adopters of CDT are involved with such schemes, the difference is statistically different at 1% level.

Table 3. Distribution of socio-demographic, socio-economic and farm characteristics of Charco-Dam Technology (CDT) adopters and non-adopters

Characteristics	Pooled (n = 528)	Adopters (n = 220)	Non-adopters (n = 308)	Mean Difference
Socio-Economic Characteristics
Male household heads	86.93	90.00	84.74	3.125†*
Households with formal education	84.47	85.00	84.09	0.080†
Mean age of household heads	38.30	39.51	37.43	−2.053^§**
Mean household size	5.11	6.70	3.98	−19.204^§***
Mean households’ off-farm incomes	1251214	1361548	1172404	−1.357^§
Individual-owned vegetable plot	47.54	47.27	47.73	0.011†
Membership to micro-finance schemes	38.83	56.36	26.30	48.840†***
Farm-Level Characteristics
Mean vegetable planted land-size	2.38	2.98	1.95	−10.313^§***
Farmers planted tomatoes	24.81	24.09	25.32	0.105†
Farmers planted cabbage	9.66	10.91	8.91	0.675†
Farmers planted sweet-pepper	27.84	35.91	22.08	12.220†***
Farmers planted scarlet-eggplant	17.99	20.00	16.56	1.030†
Farmers planted leaf-vegetables	41.48	38.18	43.83	1.687†
Use of improved seeds	78.60	80.91	76.95	1.197†
Use of chemicals	92.99	92.73	93.18	0.041†
Use of inorganic fertilizers	79.73	76.82	81.82	1.986†
Mean Labour used for farm activities	4.98	6.16	4.13	−15.823^§***
Information-Sharing Characteristics
Membership to community development groups	41.10	56.82	29.87	38.498†***
Access to extension services	95.83	95.45	96.10	0.136†
Access to radio	79.73	78.18	80.4	0.563†
Access to mobile-phone	77.27	75.45	78.57	0.710†

***, **, and * is significance at 1%; 5%, and; 10% levels; T-test values denoted by §; ${\chi }^2$ values denoted by †

Source: Research Data, 2018

Moreover, it was observed that, only three (30%) out of ten variables indicating farm-level characteristics are statistically different (i.e. land-size, those planted sweet-pepper and labour used in farming), the rest (i.e. use of inputs [seeds, chemicals and fertilizer], those planted other vegetables [tomato, cabbage and leafy vegetables]) are not statistically different. In addition, only 1 variable (Membership to community development groups) of the four information sharing characteristics examined was statistically different (at 1% level of significance). Majority of the development groups formulated in the villages or wards level are based on the agriculture, environment and social activities. The attention were given to the groups that were randomly accepting the membership, with the purpose of tracking their influence in improving members knowledge through interactions. No statistical difference observed for the two groups (adopters and non-adopters of CDT) in the; farm production variables, which include: use of improved seed, use of chemicals and use of inorganic fertilizers, as well as for the all information-sharing variables, which include: access to extension service, access to radio behaviour and access to mobile-phone (Table 3).

3.2. The diagnostic statistics of the regression models

Generally, Table 4 shows the results of probit model with instrumental variables (IVProbit). This model was having a Wald chi² of 368.62 (significant at 1% level), and Wald test of Exogeneity of 46.63 (Significant at 1% level). The IVProbit model taking the ML assumptions as a normal probit, but assume that the explanatory variables contain exogenous variable and at least one continuous endogenous variable. Using proper instrumental variable—for this case variables (NoLabourFA) and (HHinc), replaced variable (Hhsize)—override the problem of mis-measurement error observed in normal probit model. Both simple check of the Wald test of exogeneity from a “ivprobit regression” by Stock and Yogo (2002) as well as the test of IV strength by Olea and Pflueger (2013) were applied and found that the selected instrument were statistically strong. The fact that ivprobit is robust over LPM and normal probit regression, the results for IVProbit model were considered in the discussion.

Table 4. Regression results for determinants of Charco-dam technology adoption decision

Variables	Marginal Effects	Std Err	Z-value
Malehhead	0.098	0.204	0.630
Agehead	−0.008	0.006	0.192
Eduhead	−0.038	0.169	0.824
Hhsize	0.815***	0.049	0.000
Landown	−0.188	0.145	0.195
MemberFin	0.328**	0.150	0.029
UseOSW	−0.647***	0.137	0.000
Lnlandsize	0.116	0.176	0.510
Labour	−0.077	0.141	0.584
Tomato	0.453**	0.188	0.016
Cabbage	0.518*	0.275	0.059
Sweetpepper	0.004	0.192	0.982
Scarleteggplant	0.243	0.186	0.190
Leafvege	0.402**	0.171	0.019
Memberdeve	0.296**	0.141	0.036
Extser	0.068	0.182	0.709
Radio_listen	0.306*	0.161	0.058
Constant	−4.487	−7.510	0.000
No. Obs	528
Wald chi2 (17)	368.62***
Wald test of Exogeneity 46.63***
Log Pseudolikelihood	−1034.832

3.3. Determinants of Charco-dam technology adoption decision

Results in Table 4 show that household size was very significant (at 1%) and positively related with the probability of adopting CDT. The results further indicates that, one more increase in household member increases the likelihood of adopting the CDT by 0.815. Literature claims that, farming households with more family members tend to have more labour and have more chances to adopt agricultural technology than household with less family members (Kansiime et al., 2014; Kebede et al., 1990; Lambrecht et al., 2014). Likewise for the charco-dam technology (CDT), the nature of the CDT which is labour-intensive (especially during contraction of the dam, channeling the water to the dam, irrigating the crops, controlling siltation and enlarging the dam size), usually dictates the adopters to have such manpower at household level.

Most of the agricultural technologies especially those related to soil and water have accompanied by initial cost of investment (Atampugre, 2014; Yigezu et al., 2018). Thus, farmers who have substantial and consistent incomes are more likely to adopt these technologies compared to those who have low and/or inconsistent incomes (Alam, 2015). Likewise, farmer with access to credit enables him/her to have enough money to cutter for any initial or operation costs that associated with the agricultural technology opted (Abdallah, 2016; Mekonnen, 2017; Obayelu et al., 2017; Obisesani et al., 2016; Ullah et al., 2018). The findings in Table 4 are also in-line with this allegation, as individual vegetable producers in the study area who accessed micro-finance (tracked through membership to various microfinance schemes for the past three years), observed to have 0.328 likelihood of adopting CDT than those who were not members of any credit scheme, the effect is statistically significant at 5% level.

Basically, small-holder vegetable producers in the study area have two main sources of water to irrigate their vegetables; the first source is community ponds, which include open ponds, dams, wet-valleys and catchment areas that have been established and managed by the community (in this study referred as other sources of water -OSW); the second source of water is from charco-dams (CDT), which is individually established and managed at household level (the major concern of this study). Both adopters and non-adopters of CDT assumed to have an access to OSW upon adhering to the location and existing water use regulations. So, results in Table 4 show that, use of OSW is very significantly (at 1%) and negatively related with the probability of adopting CDT. Given the fact that irrigation investments (both local and improved) are cost effective, in terms of—economic, technical and environmental benefits, among others (Nissen-Petersen, 2006; Rwehumbiza, 2007), it was observed that farmers with alternative sources of water for irrigation reduce the chance of adopting CDT by 0.647.

Generally, vegetables are water demanding crops regardless of their agronomic requirements. However, the amount of water required per crop with respect to agronomic specification is varying from one vegetable-crop to another. Farmers with less access to water for irrigation would prefer less water demanding vegetables and the vice-versa holds true. Results in Table 4 show that cultivating tomato, cabbage and leafy-vegetables increase the chance of adopting CDT. Since all three vegetables are water demanding crops, producing such crops observed to increase the chances of adopting the CDT by 0.453 for tomato (at 5% level of significance), 0.518 for cabbage (at 10% level of significance) and 0.402 for leafy-vegetables (at 5% level of significance).

Essentially there is a close positively relationship between access to information and adoption agricultural technologies. Farmers who have access to agricultural information tend to have a wide knowledge and skills on various agricultural related issues, thus increase the diffusion of the various agricultural technologies and eventually increase adoption of the technologies (Baloch & Thapa, 2016; Chandio & Jiang, 2018; Maffioli et al., 2013; Nakano et al., 2018). This has also been observed, as results in Table 4 show that, access to agricultural-related information has a strong likelihood and positively influence the adoption of the CDT, as the variables associated with the acquisition and sharing of information at rural setup, i.e. membership to social and community groups is significant at 5% level.

Individuals interaction to various social and development group in the community has been proved to provide social networks, relationships and linkages that enable individuals to cooperate, coordinate, share information and resources, and eventually act collectively, which all together can increase their likelihood to adopt various agricultural related technologies (Ali et al., 2007; Hansen & Roll, 2016; Hunecke et al., 2017; Husen et al., 2017). Likewise, this study has also evaluated the status of respondents’ social capital through membership of vegetable farmers to various social and development groups in the study area and found that, membership to various social and development activities groups in the study area improves the likelihood of adopting CDT by 0.296.

On the other hand, previous studies observed that farmers who are often listening to agricultural radio programs had more chances to adopt various agricultural related interventions than those who are rarely or not listening those programs at all, simply because this behaviour let them grasp more information about agriculture that stimulate their desire to adopt various agricultural technologies (Agwu et al., 2008; Alia et al., 2013; Manda & Wozniak, 2015). The same has been observed in this study, as the results in Table 4 show that frequency of radio listening behaviour can increase the probability of CDT adoption by 0.306, this can be attributed by the fact that, most of the individual small-holder vegetable farmers in the study area who have a desire of improving their wellbeing through improving their vegetable production requires more information regarding vegetable production, this has pushed them to solicit the information to elsewhere possible including radio. The information gathered through focused group discussion reveals that most of the farmers listening to agricultural-related radio programs which focusing on weather forecasting, Rain-Water Harvesting technologies, agricultural productivity (specifically on vegetables) and market information (both input and output markets), most of these individuals have also happened to adopt CDT as well.

4. Conclusion and recommendation

Generally, the results show that household size, access to credit, choice of crop (especially tomato, cabbage and leafy-vegetables) and access to information have positive influence on adoption of CDT, while use of other sources of water for irrigation (as an alternate-e irrigation-water source) was having a negative relationship with the adoption of CDT. Having these results, it can be concluded that, given the nature of the technology, socio-economic, farm-level and information-sharing factors are all important to enhance the adoption of the technology. Thus it is suggested that, any strategies or innovation or policy geared to improve the adoption rate for such technologies should be designed or developed while considering the three key factors, i.e. socio-economic, farm-level as well as the information sharing aspects. For instance, (i) microfinance institutions should develop special loans to small-holder vegetable farmers to overcome initial investment for such technologies; (ii) Local Government Authorities, Extension Officers and agricultural development partners should use progressive farmers to instill knowledge of such technologies to fellow farmers in their development groups, as well as; (iii) using media campaigns to instill the awareness to farmers on CDT technology and related issues, such media could be local radios and social media groups in ward and village level.

Authors’ contributions

ST developed the concept, collect data, analyzed data, and draft the manuscript. RL and YM reviewed the literature and write the manuscript. PD edit and proof read the manuscript. All authors read and approved the final manuscript.

Availability of data and materials

All the data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgements

This paper benefited from the generous support individuals, groups, and institutions provided. The authors are grateful to the Environment for Development (EfD)-Initiative at Gothenburg University-Sweden and EfD-Tanzania at the School of Economics-University of Dar-es-salaam for the financial support in conducting this study. The authors also wish to record sincere appreciation to the smallholder farmers and extension staff who participated in data collections, and we thank “anonymous” reviewers for their so-called insights.

Disclosure statement

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

Additional information

Funding

The authors received no direct funding for this research.

Notes

1. www.tabora.go.tz The United Republic of Tanzania President’s Office Regional Administration and Local Government (PORALG), Tabora Regional Administrative Secretary.

Appendix A

General Information of the Used Models

Three models were considered to analyze the determinants of CDT adoption. Generally, the results in Appendix B show that, model (1) and model (2) have related findings in terms of sign, they only differ in terms of the magnitudes. While results for model (3) were different from the rest.

Model (1)—is Linear Probability Model (LPM), taking the assumptions of Ordinary Least Square (OLS), the model found to have R-square of 0.5891, which indicates that the independent variables were able to explain the adoption effect by only 58.9%.

Model (2)—is normal probit regression taking the assumptions of Maximum Likelihood (ML) while assuming no-endogeneity explanatory variables. This model was having a Waldi chi² of 185.52 (significant at 1% level), and a Pseudo R² of 0.5811, again indicating that only 58% of the variation in dependent variable are explained by independent variables. Since, this model assumed all of the explanatory variables are exogenous, hence there is a chance of mis-measurement error if otherwise (Lewbel, 2000).

Model (3) shows the results of probit model with instrumental variables (IVProbit). This model was having a Wald chi² of 368.62 (significant at 1% level), and Wald test of Exogeneity of 46.63 (Significant at 1% level). The IVProbit model taking the ML assumptions as a normal probit, but assume that the explanatory variables contains exogenous variable and at least one continuous endogenous variable. Using proper instrumental variable—for this case variables (NoLabourFA) and (HHinc), replaced variable (Hhsize)—override the problem of mis-measurement error observed in normal probit model.

Appendix B

Regression Results for; Logit Probability Mode (Model 1); Normal Probit (model 2), and; Probit with IV (Model3)

Table

Variables	Model (1) LPM		Model (2) Normal Probit		Model (3) IVProbit
	Coefficient	Std Err	Coefficient	Std Err	Coefficient	Std Err
Malehhead	0.0653	0.0422	0.018	0.040	0.098	0.204
Agehead	−0.0011	0.0013	0.000	0.001	−0.008	0.006
Eduhead	−0.0221	0.0397	−0.018	0.034	−0.038	0.169
Hhsize	0.1013***	0.0084	0.089***	0.007	0.815***	0.049
Landown	−0.0280	0.0319	−0.037	0.029	−0.188	0.145
MemberFin	0.1109***	0.0314	0.081***	0.026	0.328**	0.150
UseOSW	−0.1650***	0.0300	−0.137***	0.026	−0.647***	0.137
Lnlandsize	0.0946***	0.0318	0.092***	0.028	0.116	0.176
Labour	−0.0306	0.0298	−0.034	0.027	−0.077	0.141
Tomato	0.0784*	0.0406	0.069**	0.041	0.453**	0.188
Cabbage	0.0658	0.0559	0.038	0.056	0.518*	0.275
Sweetpepper	0.0294	0.0424	0.000	0.036	0.004	0.192
Scarleteggplant	0.0397	0.0449	0.028	0.038	0.243	0.186
Leafvege	0.0391	0.0385	0.043	0.034	0.402**	0.171
Memberdeve	0.0765**	0.0317	0.069***	0.026	0.296**	0.141
Extser	0.2731	0.0364	0.155***	0.026	0.068	0.182
Radio_listen	0.0670**	0.0310	0.069	0.030	0.306*	0.161
No. Obs	528		No. Obs	528	No. Obs	528
F(17, 510)	43.00***		Wald chi2(17)	185.51***	Wald chi2(17)	368.68***
R-Square	0.5891		Pseudo R2	0.5811	Wald test of Exogeneity (/athrho = 0)	46.63***

***Significant at 1%; **Significant at 5% and *Significant at 10% levels.

Source: Research Data, 2018