Determinants of agroforestry technology adoption in Eastern Cape Province, South Africa

Abstract

This study explores factors that affect the adoption of agroforestry (AF) technologies in ‘Tsolo’ and ‘Lusikisiki’ areas in O.R. Tambo district in the Eastern Cape Province of South Africa. It is based on empirical evidence gathered from households in the study areas. The aim and objective of the study are consistent with the strategic priorities of the South African Department of Agriculture, Forestry, and Fisheries. The study finds that agricultural farming experience, education level of the household head, a proxy variable for household wealth status, land size owned, number of livestock owned, pace of adoption of other agricultural technologies, the scale of slope of farm land, and percent of severely degraded farm land affect the adoption of AF technologies in the study areas significantly and with the higher magnitudes of the odds ratios. These variables need to be considered in all the initiatives by government and non-governmental organizations to promote AF as a strategy to realize integrated rural development in the study areas. These findings have policy implications to advance integrated sustainable rural development strategies. The local government in the study areas can use the promotion of AF technologies for multiple purposes, particularly as a tool toward rural poverty alleviation and climate change mitigation measures.

Keywords:

• adoption
• agroforestry technologies
• Eastern Cape
• logit
• South Africa

JEL Classification:

• Numbers: O33
• Q20
• Q50
• Q55

1. Introduction

Agroforestry (AF) technologies range from traditional to recent practices such as taungya,¹ home gardens, improved fallows, multipurpose trees, plantation–crop combinations, ‘silvopasture’,² shelterbelts and windbreaks, and alley cropping (Alavalapati and Mercer 2004). AF technologies have been proven to have the potential for improving productivity and the livelihoods of rural farmers (Garrity 2006; Swallow et al. 2009). However, very little effort has been made to promote AF technologies in South Africa irrespective of their vast potential in the country. The objective of this household (HH) based study is to explore the determinants of AF technology adoption in the Eastern Cape Province of South Africa. Contrary to what is known about most countries in Southern Africa, the practice of AF is not well developed in South Africa. The Eastern Cape Province was selected for the study due to its vast potential for agricultural development in general, and AF practices in particular. The rationale of the study is that a significant proportion of the South African population (close to 38%) reside in rural parts of the country, where 72% live below the poverty line (World Bank 2012). The study is aimed at the promotion of AF technologies as a means of alleviating poverty and promoting sustainable development among the rural population.

This study focuses on exploring the impact of four categories of variables on the adoption of AF technologies, namely, household preferences, household endowments, household risk and uncertainty factors, and biophysical factors. Most studies are confined to analyzing one category of those variables (see Appendix 1 for the survey of previous studies). This study is the first of its kind in South Africa analyzing four categories of the most influential factors in AF technology adoption comprehensively using an appropriate conceptual framework as depicted in section three. The study hypothesizes that these four categories of explanatory variables have an impact on the dependent variable the adoption of AF technologies in the study areas (see Appendix 2 for hypothesis, variable definitions, and expected signs). The study aims at drawing attention to the potential of AF technologies in the context of the Eastern Cape Province. This study attempts to answer the research question: what are the major socioeconomic and biophysical factors affecting the adoption of AF technologies in the Eastern Cape Province? The rest of the paper is organized as follows. Section 2 presents the literature review. Section 3 depicts the conceptual framework of the study. Section four discusses the methodology and data analysis; and section five presents results and discussion. Finally, the last section highlights the conclusion and policy implications.

2. Literature review

2.1. Socioeconomic characteristics of Eastern Cape

Despite high statistical averages in the major macroeconomic variables of South Africa, even better than many of the Organization for Economic Co-operation and Development countries, environmental challenges such as climate change and water scarcity threaten the sustainability of economic growth (OECD 2013). The challenges are much more severe in dominantly rural provinces of the country such as the Eastern Cape Province. Most of the key socioeconomic indicators are worse in the Eastern Cape Province than in the rest of South Africa. These are largely the reflections of both the homeland inheritance and the generally poorer status of the province. More than 55% of the working age population in the province has no permanent monthly income (see Figure 1). Forty percent of the land area of the province comprises the former homeland areas of Transkei and Ciskei.

Figure 1. Percentage individual monthly income of working age populations in Eastern Cape. Source: Own computation on the basis of Statistics South Africa (2012).

The economy of the province is based to a large extent on subsistence agriculture and on income from pensioners and migrant laborers working outside the region. Recently, the Eastern Cape population was estimated at 6,562,053, an increase of 283,402 persons or 4.5% since 2001 (Statistics South Africa 2012). Agricultural HHs in general earn less than their nonagricultural counterparts in the province. Agricultural HHs often reside in rural areas and as a result, the national government should place strong emphasis on smallholder farmers to curb massive rural poverty in the Eastern Cape Province in particular and in the country in general (Aliber and Hall 2012).

2.2. AF in South Africa

AF was introduced in South Africa as early as 1887 (Nair 1993). Unlike in most Southern African countries, AF practices are not well established in South Africa. Southern African countries such as Malawi, Namibia, Tanzania, Zambia, and Zimbabwe have benefited from the Southern African Development Community (SADC)-International Center for Research in Agroforestry (ICRAF) Zambezi Basin AF Project since the mid-1990s. South Africa has not been collaborating in such institutional partnerships and national efforts toward promoting AF in smallholder farming systems. However, in the tree-rich savannah lands of South Africa, such as parts of the Eastern Cape, Northern Natal, the Lowveld and Bushveld in the Northern Province, and the Kalahari where livestock farming is practiced, trees are protected for the production of additional fodder for the drought the season, as a source of fencing material and firewood, for stabilizing soil, for providing shade, and for general environmental conservation purposes (FAO 2002).

The major constraint for raising livestock in the summer rainfall areas of South Africa is the shortage of fodder available to livestock during winter. Since in winter there is limited rainfall and lower temperatures, grass growth is restricted. Conversely, increased rainfall and temperature in summer should create more favorable conditions. A possible solution to the problem of winter-feeding has been learned from the habits of goats and their ability to browse and utilize the leaf material of trees. Trees, once established, need little ongoing care and provide a ready source of plant material during winter months. Exotic tree species that are used for fodder in South Africa include Prosopis species, such as Ceratonia siliqua, Sesbania sesban, Leucaena, and Chamaecytisus palmensis. These trees are mostly legumes and perform a number of valuable functions apart from being fodder crops. In South Africa, there are a number of indigenous species that are used as fodder trees. There are also a wide variety of Acacia species that can benefit livestock. (see Appendix 3 for the list of AF trees suitable for South African climatic conditions). Note that this is not an exhaustive list of AF tree species useful to South Africa. Everson et al. (2009) conducted an on-farm study in the Upper Thukela region of KwaZulu-Natal to determine dry matter production for fodder tree species and their effect on soil water and maize production.

The study by Mafongoya et al. (2006) has also confirmed that AF is one of the options appropriate and available to smallholder farmers to replenish soil fertility in Southern Africa. In this study the AF option is ranked highest by farmers among the options available in contributing to soil fertility, but the lowest in terms of ease of adoption by farmers. Therefore, there is a need to research the issues of adoption and scaling up of AF technologies in the Republic of South Africa.

2.3. Socioeconomic and environmental benefits of AF

Despite the increasing realization of AF as an environmentally and economically suitable land-use practice, the adoption of AF technologies is very slow across all sub-Saharan African (SSA) countries (Franzel and Scherr 2002; Mafongoya et al. 2006). Because of the three major components constituting the AF system, AF technologies comply with environmental guidelines while enabling subsistence and small-scale farmers to improve their yield per plot of land. Broadly defined, AF is ‘a land use that involves deliberate retention, introduction, or mixture of trees or other woody perennials in crop/animal production fields to benefit from the resultant ecological and economic interactions’ (MacDicken and Vergara 1990). AF involves the integration of trees into farming systems in ways that create an agroecosystem succession, similar to that in natural systems (MEA 2005).³ AF is a promising land-use practice to maintain or increase agricultural productivity while preserving or improving agricultural land fertility. It does not convert agricultural land to forests, but rather leaves land in production agriculture, while integrating trees into farm and ranch operations to accomplish economic, environmental, and social goals.

Recent studies show that AF practices in Africa have a huge potential to sequester atmospheric carbon dioxide (Luedeling et al. 2011). AF practices have considerable potential in helping solve some of Africa's main land-use problems (Cooper et al. 1996; FAO 2013; Sanchez 1995) through the provision of a wide range of tree products for domestic use or sale (Franzel et al. 2001). Swallow et al. (2009) argue that AF farming can contribute toward achieving the Millennium Development Goals in African countries.

AF plays a significant role in increasing agricultural productivity by nutrient recycling, reducing soil erosion, improving soil fertility and by enhancing farm income compared to conventional crop production (Kang and Akinnifesi 2000; Neupane and Thapa 2001; Neupane et al. 2002). AF can also potentially reduce deforestation while increasing food, fodder, and fuel wood production (Neupane and Thapa 2001; Neupane et al. 2002). Benefits that accrue from the usage of AF include food and nutrition security, increased income and assets, and improved land management (Garrity 2006); it also creates environmental and management synergies (Race 2009).

There is increasing evidence that the potential of AF to reduce poverty is real and can be put to effective use in the Poverty Reduction Strategies of many countries in Africa. In forest-scarce countries, AF has expanded greatly on small farms. In Kenya and Ethiopia, for example, farms account for most timber and pole production. In AF systems, the cost of tree production may be lower due to joint production with crops and livestock. Trees have a positive effect on the incomes of associated crops, as in the case of use as windbreaks (Jama and Zeila 2005).

When we consider the environmental benefits of AF, we recall great global events such as the 1992 Rio Earth Summit. Agenda 21, the blueprint for action in the twenty-first century adopted by world leaders meeting at the Summit, identifies AF as one way of rehabilitating the degraded dry lands of the world. AF, one of the several approaches for improving land use, is also frequently invoked as an answer to shortages of fuel wood, cash income, animal fodder, and building materials in SSA (Rocheleau et al. 1988). The environmental benefits of AF include soil erosion control (Young 1989), improvement of soil quality through increased nitrogen input, improvement of water dynamics (Phiri et al. 2004), and increased activity of soil biota (Sileshi and Mafongoya 2006). AF systems such as woodlots do supply fuel wood and can therefore alleviate the demand from natural forests and hence reduce deforestation (Sileshi et al. 2007). They have also shown that they can sequester carbon, though at different rates depending on the species used and management regimes and systems (Kaonga and Coleman 2008; Kaonga and Bayliss 2009; Sileshi et al. 2007).

AF systems have also demonstrated their ability to conserve biodiversity and suppress insect pests and weeds (Sileshi et al. 2005; Sileshi and Mafongoya 2006; Sileshi et al. 2007) better than monoculture agricultural systems. Mafongoya et al. (2008) discuss some of the technically feasible and financially affordable technologies which are appropriate and available to farmers. For further details on the socioeconomic challenges and constraints that limit the adoption of these options, see Mafongoya et al. (2008). In the next section, we discuss the conceptual framework of the study considering the points highlighted in this and preceding sections.

3. Conceptual framework of the study

Different frameworks and approaches have been used for the analysis of adoption of AF technologies. Biot et al. (1995) grouped these approaches into three major types: top-down interventions, populist or farmer-first, and neo-liberal approaches. Building from the farmer-first and sustainable livelihood principles but extending and incorporating important elements from various theories and practical realities, Shiferaw, Okello, and Reddy (2009) have developed a broader conceptual framework for the analysis of factors conditioning the adoption and adaptation of smallholder natural resource management technologies in general. However, such a framework is too broad and complex to analyze the adoption behavior and institutional setup of AF technologies concurrently. Given the focus of this study, the conceptual framework developed by Kant and Lehrer (2004) is more appropriate.

For the purpose of this study, the conceptual framework shown in Figure 2 is modified and adapted from Kant and Lehrer (2004). The conceptual framework developed for this study incorporates the adoption of AF technologies as a core result of the interaction of factors affecting the AF adoption and institutional setup. In our conceptual framework shown in Figure 2, the institutional aspect incorporates the integrated sequential nature of the AF production process (physical and socioeconomic) with micro, macro, formal, and informal institutions, and the requirements for social, economic, and ecological sustainability as the evaluative criteria (for details, see Alavalapati and Mercer 2004). In our conceptual framework, the explanatory variables in the adoption decision of AF technologies are depicted on the left-hand side of the figure. Institutional factors are depicted on the right-hand side along with the physical and technical attributes of resources and actions and the interaction of external and internal agents (i.e. resource uses and users and stakeholders) in the middle. However, in this study we focus only on the socioeconomic and biophysical factors, putting aside institutional analysis for future study.

Figure 2. Conceptual frameworks for the analysis of institutional and non- institutional factors in the adoption of natural resources management.⁵ Source: Modified and adapted from Kant and Lehrer (2004).

4. Methodology

4.1. Data and sampling method

Quantitative data were collected from Tsolo and Lusikisiki Magisterial Districts of the Eastern Cape Province using a pretested, validated, and standardized questionnaire. Open questions were included in the questionnaire so that if triangulation was needed, then additional information was available. The study employed triangulation to determine both the accuracy and the authenticity of sources.

The design of the study was cross-sectional, descriptive, and co-relational. The sample size of study was 300 HHs. Simple random sampling was used as the sampling technique. Sampling frames were obtained from Statistics South Africa for the purpose of selecting eligible HHs for the survey. Quantitative data analysis was performed by using statistical methods, such as cross-tab analysis and logit regression analysis. Table 1 presents the descriptive statistics of the variables included in the data analysis.

Table 1. Descriptive statistics of variables included in the data analysis.

Explanatory variable code	Variable description	Mean	Std. Dev.
Category 1: Household preference variables
GenderHH	Gender of the household head	0.47	0.500
EducationHH	Education level of the household head	5.34	3.050
AgeHH	Age of household head	59.45	13.581
NHMR	Number of houses with modern roof	2.23	1.386
GFE	General farming experience	20.18	14.4735
TTTP	Tendency toward tree planting (1 = high, 0 = otherwise)	0.73	0.446
Category 2: Household resource endowment variables
HHS	Household size	7.14	3.141
LS	Land size owned	2.343	1.8899
NoL	Number of livestock	16.17	22.438
IFAA	Income from agricultural activities	0.33	0.497
FC	Financial capital	0.48	0.500
Category 3: Household risk and uncertainty factors (variables)
LT	Land tenure (1 = private, 0 = otherwise)	0.69	0.505
NOL	Number of livestock	16.17	22.438
SMP	Strong market participation	0.23	0.422
OFI	Off-farm income	0.13	0.560
CR	Credit risk	0.64	0.480
POAFA	Pace of adoption of other agricultural technologies (1 = early adopter, 0 = late)	0.37	0.484
Category 4: Biophysical factors (variables)
SFS	Soil fertility status (1 = fertile, 0 = otherwise)	0.67	0.472
SOFL	Slope of farm Land (1 = high slope, 0 = plain)	0.42	0.495
DFEO	Distance from extension office	10.58	3.446
DFNMP	Distance from nearest market place	10.61	3.756
EOPSDFL	Estimate of percent of severely degraded farm land (1 = severely degraded, 0 = otherwise)	15.14	11.488
HHs adopted AF	217
Valid N	300
Dependent variable	Adoption of AF technologies: 1 if ‘yes', 0 otherwise – in all the categories

Source: Authors from survey data.

From the total respondents, more than 72% of them have been practicing AF in their farms, while 28% are not practicing AF. Such a difference between adopters and non-adopters of AF is because of the intentional inclusion of more farmers practicing AF in the sample for the purpose of exploring factors affecting AF practices in the two districts.

4.2. Logit model specification

The modeling approach considers adoption as a dichotomous dependent variable, which takes ‘1' if adoption is present and ‘0' otherwise. The model produced in logistic regression is nonlinear and the outcome variable, Y, is the probability of having one outcome or another based on a nonlinear function of the best linear combination of predictors, with two outcomes. In binary regression models goodness of fit (R² values) are not important; the important feature is the expected signs of the regression coefficients and their statistical and/or practical significance. Therefore, the interpretation focuses on statistical significance, direction of the regression coefficients (either positive or negative), and the odds ratios.

As specified in Agresti and Barbara (2009) and Peng and So (2002) the simple logistic regression model has the form: (1) When we take the antilog on both sides of Equation (1), we derive the equation to predict the probability of the occurrence of the outcome of interest as shown in Equation (2): (2) where ‘’ is the probability of the outcome of interest (Y = 1); ‘’ is the Y intercept (constant of the equation);

’ represents the regression coefficients of the explanatory variables (i.e. vector of coefficients to be estimated); represents a set of predictors, and ‘’ is the base of the system of the natural logarithms. Taking the log of Equation (2) we have the following logit model for estimating coefficients: (3) Finally, we estimated Equation (3) using statistical software to find the best linear combination of predictors to maximize the likelihood of obtaining the observed outcome frequencies. The estimation results and economic interpretations are presented in the next section. Interpretations are given in terms of odds ratios and not in terms of marginal effects. Marginal effects are suitable for linear probability models, whereas in the case of binary response models odds ratios give more intuitive meaning. If the odds ratio, Exp (β), is greater than 1, we interpret it as the odds are ‘exp (β)' times larger. If the odds ratio is less than 1, we take it as the odds are ‘exp (β)' times smaller, holding all other variables constant (Gujarati 2004; Long 1997; Menard 2001).

5. Results and discussions

In this section the results from the empirical analyses for adoption of AF are discussed. Factors affecting the overall AF technologies in the study area are analyzed by categorizing the factors into four categories. The categories of explanatory factors affecting the adoption of AF technologies in the study are household preferences, household resource endowments, risk and uncertainty factors, and biophysical factors. Each of these categories of explanatory variables are analyzed and discussed in the following.

5.1. Logistic regression result of HH preference

Table 2 summarizes the results of the logistic regression of HH preference variables on the adoption of AF technologies. The Omnibus test of model coefficients indicates that the model containing all the predictors was statistically significant. As shown in Table 2, except age of HH head and tendency toward tree planting the rest of the explanatory variables significantly affect a HH's adoption of AF technologies. The last column in all the result tables shows Exp(β) (the exponential coefficient), the maximum likelihood estimate of the odds ratio, which we use for interpretation purposes. The odds ratio of the gender of HH head is 0.535 with a negative coefficient, which implies that being a male-headed family reduces the likelihood of adopting AF technologies by 0.535 units at 5% level of significance, holding other factors constant. The education level of the HH head has a positive coefficient and an odds ratio of 1.142, which implies that holding other factors constant, more educated HH heads have more than one times likelihood of adopting AF technologies than uneducated or less educated HH heads in the study areas at 10% level of significance. The number of houses with a modern roof, the proxy variable for the wealth of the HH, has a positive coefficient with an odds ratio of 1.037, which implies that wealthy HHs will adopt AF technologies by more than one time than poor HHs at 1% level of significance, holding other things constant. The other significant variable in the model is general farming experience with a positive coefficient and an odds ratio of 4.488. Holding the other explanatory variables constant in the model, HHs with more general farming experience will adopt AF technologies more than four times than those with less general farming experience. Age of HH head and tendency toward tree planting are the two explanatory variables with insignificant relationships with the dependent variable; however, both these variables have positive coefficients. In this model the only variable with a negative coefficient is the gender of the HH head (1 = male, 0 = female). Since the interpretation goes with a nominal value of 1, male-headed HHs disfavor AF adoption. Such HHs may favor cash crops in a monoculture type of production with short-term monetary rewards.

Table 2. Logistic regression estimation of the household preferences for the adoption of AF technologies.

Variable name	β	Stad. error	Wald	Df	Sig.	Exp(β)
Gender of the household head	−0.625**	.306	4.167	1	.041	.535
Education level of the household head	0.132*	.052	6.371	1	.010	1.142
Age of household head	0.105	.129	.662	1	.416	1.110
Number of houses with modern roof (proxy variable for wealth status)	0.036***	.012	8.506	1	.004	1.037
General farming experience	1.501***	.306	24.051	1	.000	4.488
Tendency toward tree planting	0.010	.012	.654	1	.419	1.010
Constant	−1.844	.883	4.363	1	.037	0.158
−2 Log likelihood			294.480
Cox & Snell R^2			.180
Nagelkerke R^2			.259
Omnibus tests of model coefficients
• Chi-square			59.387
• Df			6
• Sig.			.000
% Correct predictions			79.3

Source: Author, computed from survey data.

*Significant at 10% level of significance (p < 0.10).

**Significant at 5% level of significance (p < 0.05).

***Significant at 1% level of significance (p < 0.01).

5.2. Logistic regression result of HH resource endowments

As shown in Table 3, five variables of HH endowments are regressed with the dependent variable adoption of AF technologies. All of the variables are scale variables except the dummy variable for financial capital. The dummy variable for financial capital is obtained from the respondents' reply for the request on access to credit services. The Omnibus test of model coefficients indicates that the model containing all the predictors is statistically significant at 5% level of significance and gives 72% correct predictions.

Table 3. Logistic regression estimation of the household endowment for the adoption of AF technologies.

Variable name	β	Stad. error	Wald	df	Sig.	Exp(β)
Household size	–.032	.042	.570	1	.450	.969
Land size owned	–.134*	.075	3.209	1	.073	.874
Number of livestock	.011	.008	2.210	1	.137	1.011
Income from agricultural activities	.260	.300	.749	1	.387	1.297
Financial capital	.345	.274	1.590	1	.207	1.412
Constant	1.108	.363	9.316	1	.002	3.029
−2 Log likelihood				345.571
Cox & Snell R^2				.027
Nagelkerke R^2				.039
Omnibus tests of model coefficients
• Chi-square				8.295
• Df				5
• Sig				0.05
% Correct predictions				72.0

Source: Author, computed from survey data.*Significant at 10% level of significance (p < 0.10).***Significant at 1% level of significance (p < 0.01).

In this model, as shown in Table 3, HH size and land size owned by the HHs have negative coefficients and the odds ratios are below unity, which implies that the larger the values of these variables, the lower the adoption of AF technologies. The rest of the variables included in the model, namely, number of livestock owned, income from agricultural activities, and financial capital, has positive coefficients with larger odds ratios. These values imply that these variables have a positive effect on the adoption of AF practices in the study areas.

5.3. Logistic regression result of HH risk and uncertainty factors

HH risk and uncertainty variables are described below in Table 4. In this section we regress our logit model using the dependent variable, AF adoption, over HH risk and uncertainty variables is robust enough to analyze the relationship between the dependent and explanatory variables. However, as shown in Table 4 only two explanatory variables i.e. number of livestock in the HH and the pace (speed) of adoption, are statistically significant at 10% and 1% level of significance respectively.

Table 4. Logistic regression estimation of the household risk and uncertainty factors for the adoption of AF technologies.

Variable name	β	Stad. error	Wald	df	Sig.	Exp(β)
Land tenure	.039	.280	.019	1	.889	1.040
Number of livestock	.012*	.008	2.628	1	.100	1.013
Strong market participation	.213	.330	.416	1	.519	1.237
Off-farm income	.082	.232	.125	1	.724	1.085
Credit risk	.371	.285	1.689	1	.194	1.449
Pace of adoption of other agricultural technologies	1.631***	.345	22.326	1	.000	5.107
Constant	−.004	.332	.000	1	.992	.996
−2 Log likelihood				321.357
Cox & Snell R^2				.103
Nagelkerke R^2				.148
Omnibus tests of model coefficients
• Chi-square				32.509
• Df				6
• Sig.				.000
% Correct predictions				72.7

Source: Author, computed from survey data.*Significant at 10% level of significance (p < 0.10).***Significant at 1% level of significance (p < 0.01).

The larger the number of the livestock, the higher will be the likelihood of adoption of AF technologies by the sampled HH, holding other things constant. As shown in Table 4, HHs with the larger number of livestock will adopt by more than 1.013 times than the HH with lesser number of livestock. Similarly, the odds ratio of being an early adopter is very high with Exp(B) = 5.107. This odds ratio implies that holding other things constant, early adopters of other agricultural technologies adopt AF technologies by more than five times than the late adopters. All the rest of the explanatory variables also positively affect the adoption of AF practices by more than one time. Land tenure is the most controversial variable in most adoption studies. In this study, land tenure type is statistically insignificant and yet private ownership of land favors adoption of AF practices in the study areas.

5.4. Logistic regression result of biophysical factors

Most socioeconomic studies on AF adoption overlook the effect of biophysical factors in the adoption of AF technologies. As indicated in Table 5, the Omnibus test of model coefficients of this model reveal that the model for logistic regression of biophysical factors is robust enough at 1% level of significance and with more than 70% correct predictions.

Table 5. Logistic regression estimation of the biophysical factors for the adoption of AF technologies.

Variable name	β	Stad. error	Wald	df	Sig.	Exp(β)
Soil fertility status	–.425	.304	1.945	1	.163	.654
Slope of farm land	2.335***	.433	29.029	1	.000	10.329
Distance from extension office	–.195	.218	.803	1	.370	.822
Distance from nearest market place	.096	.193	.249	1	.618	1.101
Percent of severely degraded farm land	.042**	.019	5.038	1	.025	1.043
Constant	.996	.725	1.887	1	.170	2.707
−2 Log likelihood				303.033
Cox & Snell R^2				.156
Nagelkerke R^2				.225
Omnibus tests of model coefficients
• Chi-square				50.834
• Df				5
• Sig.				.000
% Correct predictions				70.7

Source: Author, computed from survey data.**Significant at 5% level of significance (p < 0.05).***Significant at 1% level of significance (p < 0.01).

In this analysis two of the explanatory variables significantly positively affect the adoption of AF technologies in the study areas. The first variable is the slope of a HH's farm land. The higher the slope of the HH's farm lands, the higher the likelihood of adopting AF technologies. In this case HHs with sloping farm lands adopt AF technologies 10 times more than those with plain lands at 1% level of significance. The second variable is the severity of a HH's farm land degradation. The odds ratio of this variable is 1.043, which is significant at 5% level of significance. Holding the rest of the explanatory variables constant, the model predicts that HHs with severely degraded farm land adopt AF technologies more than one time than those with less severely degraded farm lands. From the direction and magnitude of impact of these two variables, we can deduce that the respondents with slopy and heavily degraded land are aware of the positive role of AF tree planting in soil and water conservation.

In this model, good soil fertility status and distance from the extension office have a decreasing effect on household AF adoption. On the contrary, the distance from the nearest market place positively affects the likelihood of AF technology adoption. However, all of these three variables are not statistically significant. In this model, distance from the nearest market place favorably affects the adoption of AF technologies. It means the higher the distance between the farm land and the market place, the more will be the maximum likelihood of AF technology adoption. This result seems paradoxical. However, this seemingly inconsistent result can be related to the fact that AF products and services in the study areas are not monetized and are not marketable.

6. Conclusion and policy implications

This study has identified socioeconomic and biophysical factors affecting the adoption of AF technologies in the study areas. Groups of variables falling in each category were rigorously analyzed using a logit regression model. The study finds that there are variables that significantly affect the maximum likelihood of AF technology adoption in the study areas. These variables are gender of the HH head, education level of the HH head, number of houses with a modern roof (a proxy variable for wealth), farming experience, farm land size owned by farmers, number of livestock owned, pace of adoption of other agricultural technologies, slope of farm land, and percent of severely degraded farm. These variables need to be considered in all the initiatives by government and non-governmental organizations to promote AF in the study areas and in other parts of the country.

The socioeconomic challenges facing South Africa are veiled by high statistical averages. However, there are multifaceted challenges in rural parts of the country. As one of the land-based economic development strategies, AF can contribute to sustainable rural development with wider positive impacts on the economy of rural HHs. However, AF practices in the country are still under traditional levels of operation. To include scientific inputs on the existing AF practices from academic and research institutions, the Department of Agriculture and Forestry (DAFF) should focus on assisting farmers and extension workers with cutting-edge scientific practices and information on AF technologies. Furthermore, DAFF should facilitate financial credit services and incentive schemes for those farmers who have the experience and willingness to promote AF practices in the study areas and across the country. Improving marketing and adding value to raw products are also critical for enhancing the livelihoods of AF farmers. Community-based institutional mechanisms are needed to help farmers acquire information and business skills.

The lack of high-quality AF tree germplasm has long been recognized as one of the major challenges to widespread adoption of AF in Southern Africa (Nyoka et al. 2011). Supply of AF tree seedlings and germplasm can do better in prompting AF technology adoption in the study areas and across the country. Training should be given to farmers on tree management practices (such as pruning, thinning, and coppicing) and on individual tree and stand manipulation to reduce adverse ecological interaction with agricultural crop components of the system.

The Government of South Africa as part of its Green Economy initiative can use the promotion of AF projects for multiple purposes, particularly as a tool toward rural poverty alleviation and climate change adaptation measures. In line with this, the Government can incentivize farmers to introduce AF in their farming system as part of the Kyoto Protocol which allows for reduction in carbon emission through forest/agroforest-based carbon sequestration projects (UNFCCC 2002). Analyzing the actual potential, feasibility, and profitability of AF establishments in rural parts of South Africa is a quest for future research.

Acknowledgments

The authors would like to express their sincere gratitude to Dr Gilberthorpe, Editor in Chief, Development Studies Research, and the anonymous reviewers for the comments on the earlier draft of the manuscript that led to significant improvements in the current version. We also extend our appreciation to Tsolo Agricultural and Rural Development Institute in the Eastern Cape for the generous logistic support during data collection at Tsolo and Lusikisiki. We would also like to thank Statistics South Africa for providing us with the secondary data and sampling frames used in this study.

Funding

The first author would like to thank the National Research Foundation (NRF) of South Africa for the financial assistance to carry out this study.

Notes

1. Taungya agroforestry is when agricultural crops are grown during the early stages of forest plantation establishment.

2. Silvopasture (Latin, silva- forest) is the practice of combining forestry and grazing of domesticated animals in a mutually beneficial way. Advantages of a properly managed silvopasture operation are enhanced soil protection and increased long-term income due to the simultaneous production of trees and grazing animals. Perhaps the oldest agroforestry systems used in the temperate regions of the world, silvopastoral systems are characterized by integrating trees with forage and livestock production.

3. MEA stands for Millennium Ecosystem Assessment.

4. In all the hypotheses stated earlier, the dependent variable is adoption of AF technologies: if ‘yes' 1, if ‘no' 0.

5. The term natural resources management (NRM) refers to the sustainable utilization of major natural resources, such as land, water, air, minerals, forests, fisheries, and wild flora and fauna.

Appendix-1. Summary of empirical ex-post AF adoption studies.

Selected study	Variable included	Proxy variables used	Region
Shively (2001)	Risk and uncertainty	• Tenure security	Sumatra, Indonesia, Philippines
Pannell (2003)	• Reduce adoption	• Consumption risks (proxy by farm size)
Marra et al. (2003)	• Rarely studied	• Experience
	• Significant in 71% of studies
Mercer and Pattanayak (2003)	Household preferences	• Education	Kenya, Eastern Zambia
Scherr (1995)	• Significant in 41% of studies	• Age
Gladwin et al. (2002)	• Assumed heterogeneous	• Gender (significant)
		• Sociocultural status
Pattanayak et al. (2003)	Resource endowments	• Labor	Survey from many regions
	• Significant in 60% of studies	• Income (significant)
	• Better-off HHs are early adopters	• Land size
		• Livestock
		• Savings and credit
Hyde and Amacher (2000)	Incentives	• Input/output price*	Asia
	• Significant in 78% of studies	• Market availability
	• Studies suggest for future research	• Expected income change
		• Transportation costs
Simmons, Walker, and Wood (2002)	Institutions	• Land tenure	Brazil and Panama
Kant and Lehrer (2004)	• More important than household preference variables	• Property right
	• Not discussed since Kant and Lehrer (2004) suggest a framework for future evaluation	• Land and tree management
		• Market for AF outputs
		• Credit constraints
		• Availability of information
		• Availability of extension services

Source: Authors

Appendix-2. Hypotheses, variable definition, and expected signs.

Hypothesis 1^4	Null hypothesis (Ho)	Alternative hypothesis (H1)	Independent variables	Expected signs
	Household preference variables have a positive statistically significant impact on the adoption of AF technologies	Household preference variables do not have a statistically significant impact on the adoption of AF technologies	Gender of the household head	Positive
			Education level of the household head	Positive
			Age of household head	Positive
			Number of houses with modern roof	Positive
			Farming experience	Positive
			Tendency toward tree planting (1 =  high, 0 = otherwise)	Positive
Hypothesis 2	Household endowment variables have a statistically significant impact on the adoption of AF technologies	Household endowment variables do not have a statistically significant impact on the adoption of AF technologies	Household size	Positive
			Land size owned	Negative
			Number of livestock	Positive
			Income from agricultural activities	Positive
			Financial capital	Positive
Hypothesis 3	Household risk and uncertainty factors do not have as such a statistically significant impact on the adoption of AF technologies	Household risk and uncertainty factors have as such a statistically significant impact on the adoption of AF technologies	Land tenure	Negative
			Number of livestock	Positive
			Strong market participation	Positive
			Off-farm income	Negative
			Credit risk	Negative
			Pace of adoption of other agricultural technologies	Positive
Hypothesis 4	Biophysical factors do not have as such a statistically significant impact on the adoption of AF technologies	Biophysical factors have a statistically significant impact on the adoption of AF technologies	Soil fertility status	Negative
			Slope of farm land	Positive
			Distance from extension Office	Negative
			Distance from nearest market place	Negative
			Percent of severely degraded farm land	Positive

Source: Authors

Note: Based on binary logistic regression analysis, the hypotheses are tested for validity at 1%, 5%, and 10% level of significance.

Appendix-3. Appendix-3-Some of the AF trees suitable for South Africa.

Scientific name	Local name
Acacia albida	Ana tree
Acacia erioloba	Camel thorn
Acacia tortilis subsp. Spirocarpa	Hairy umbrella thorn
Bauhinia petersiana subsp. macrantha.	Coffee neat's foot
Colophospermum mopane	Mopane
Euclea species	Guarri trees
Olea europaea subsp. africana	Wild olive
Acacia sieberana var woodii.	Paper bark thorn
Boscia species	Shepherd's trees
Combretum species	Bush willows
Grewia species	Wild raisin bushes
Portulacaria afra	Elephant bush
Rhus viminalis	White karree
Sclerocarya birrea subsp. caffra.	Marula
Tarchonanthus camphorates	Wild camphor bush
Terminalia species	Cluster-leaf trees
Ziziphus mucronata subsp. mucronata	Buffalo thorn

Source: Underwood (1993).