3.5 Estimation strategy
3.5.1 The household-level model
We estimate the determinants of farmers’ decision to invest in agroforesty5, and in particular the effect of the land titling policy on adoption incentives. Since aggre- gating different technologies that imply different costs and benefits over time may be misleading, we focus on agroforestry which is one of the most widely known measures against soil erosion in the study area and is also perceived to be one of the most effective (cf. Section3.6.2).
The investigation of adoption determinants in a population where the diffusion of innovation is incomplete may lead to biased estimates (Diagne and Demont, 2007). Selection bias arises when exposed and unexposed farmers differ in their propensity to adopt the technology6. This may be the case for at least two reasons. First, knowledge acquisition is part of the farmers’ adoption decision and therefore endogenous and, second, for efficiency reasons agricultural extension may especially target farmers or communities with a high innovative capacity. In Section 3.6.2 we show that knowl- edge diffusion for agroforestry is incomplete in the study area suggesting the use of a selection bias correction model. Our problem can be written as follows:
y1i = ( 1hβX1i+ui >0 i if y2i = 1 0 otherwise (3.1) y2i = 1 h γX2i+vi>0 i ∀i∈[1, N] (3.2)
where N is the total population; y1i and y2i are binary dependent variables indi- cating the adoption and knowledge status of the ith household respectively; X1i and
X2i are vectors of regressors;ui and vi are the error terms, we assume that they are jointly bivariate normally distributed. The covariance matrix is:
cov(ui, vi) =
σ2
u ρ
ρ σv2 !
We use a Heckman full maximum likelihood procedure to jointly estimate the probability of knowing and adopting the technology and control for selection bias
(Heckman,1979). The model predicts household’s probability to adopt and maintain
5
“Agroforestry is a collective name for land-use systems in which woody perennials are deliberately grown on the same piece of land as agricultural crops and/or animals”(Lundgren, 1982). By agro- forestry, we refer to a cultivation technique consisting in planting trees and/or shrubs on a cultivated land as a way to limit soil erosion and improve soil fertility. The plants mostly used in the study area are wild tamarind (leucaena leucocephola), teak trees and pine trees.
6
In this context selection bias is also termed exposure bias in the literature, e.g. by Diagne and Demont(2007)
agroforestry on at least one of its plots conditional on variables X1i and on knowing agroforestry7 (Table3.1).
Table 3.1summarizes the explanatory variables (contained in X1i and X2i). Fol- lowing literature on knowledge acquisition and learning (Feder and Slade,1984;Foster
and Rosenzweig,1995;Conley and Udry,2001), we expect that information access is
closely linked to education, social capital, the possession of communication assets, access to the agricultural extension service, and income. The social capital variable measures how well the household is connected with mass organizations8 in its village by assessing how easily help is obtained if necessary, which is referred to as vertical connections in the following. In addition, we include a variable on households’ par- ticipation in the farmer union as a measure of horizontal social capital. The daily per-capita expenditure variable is used as a proxy for wealth9. Other variables control for access to agricultural extension, human capital, and possession of communication assets.
Among the regressors of adoption, control variables account for major household characteristics (the number of active members, education, age of the household head, and wealth level), soil, and farm characteristics, as well as geographic location. In addition, we include a regressor indicating whether material support was received by the household to implement agroforestry. Material support includes labour, in-kind inputs (seeds, seedlings or fertilizer, for instance) or cash support. Such support has been provided either by governmental or non-governmental organizations to en- courage certain farmers to adopt agroforestry. Hereby, several goals may have been pursued, and different types of households targeted: firstly, the focus may have been on farmers with a low investment capacity; secondly, extension organizations may have targeted influential or exemplary farmers as a way to disseminate a technology (using a demonstration plot); finally, it may also have been intended to enforce adop- tion in areas of strategic importance, such as easily visible locations close to the main road10. However, statistical tests show no systematic differences between supported and unsupported households regarding potential influencing factors of adoption (hu- man, social, and financial capital). We therefore conclude that the attribution of support was random regarding characteristics that also influence adoption and that, consequently, we do not face an endogeneity problem.
7
The probabilityP(y1i= 1|y2i= 1, X1i)is derived inWooldridge(2002a, pp.477-78 and 570-71).
8
In Vietnam, the mass organizations play a crucial role and are present at all administrative levels (from the village to the state). They are composed of six unions representing women, farmers, veteran, elderly, youth and the fatherland front union. In addition to participating in major village decisions, these organizations carry out multiple tasks: from extension agents to rural bank staffs.
9The survey contained an extensive module on households’ food and non-food expenditures, asset
values, and remittances received, from which this variable is calculated.
10It was mentioned that some farmers having their plot located close to the National road had
been strongly encouraged to implement hedgerows on their field so as to create positive impressions on officials and visitors passing by the area.
3.5. Estimation strategy 28
We hypothesize that improved access to credit is conducive to the adoption of agroforestry based SCT because it relaxes liquidity and/or consumption constraints, and reduces farmers discount rates. This will lead to a higher value being attached to benefits from reduced soil erosion that accrue in the futurePender(1996);Holden et al. (1998). We use a binary variable indicating whether a household is credit constrained on the formal credit market. FollowingZeller(1994) we consider farmers to be credit constrained if they did not apply for credit for fear of rejection or if they applied for a loan but were partially or fully rejected by the lenders11.
The effect of the land titling policy on adoption incentives is captured by five variables. A first variable measures households’ share of upland area operated under a land title. A positive and significant coefficient would indicate that land title is perceived as a guarantee of tenure security thus encouraging farmers to engage in soil conservation. Apart from being registered and operated under a title, the land can be (i) leased from private households for a defined period of time and/or a fixed payment; (ii) lent or given by private households with no payment and for an undefined time period; (iii) leased or borrowed from the village fund land; or (iv) cultivated without agreement or informally purchased (Table3.2). With this variable on land title, we test our main hypothesis - that tenure is perceived to be more secure when land is operated under a title than under any of the other tenurial arrangements. Several empirical studies have found evidence that tenure security may be endogenous to investment, as farmers may undertake certain investments to secure tenure and obtain land titles
(Besley,1995; Place and Swallow,2000;Brasselle et al.,2002). In our case, the fact
that land titles have been distributed to all households at a certain point in time (cf. Section 3.6.1and Table 3.2) excludes this risk of endogeneity, but the correlation of the land title variable with unobserved factors of adoption is might be a source of endogeneity. We conduct some test, presented in the Appendix of this paper12.
As outlined in Section 3.6.1, the implementation of the land policy in the study area has resulted in successive reallocations, and a majority of farmers expect further reallocations to take place before the end of the use right term. While the issuance of a land title was supposed to empower farmers as decision makers over the use of their land, the successive reallocations may have sent the contradictory signal that the state remains the primary decision maker over land issues. We include four variables to capture these effects: (i) a dummy variable indicating whether the household has experienced a reallocation on its upland plots, (ii) the share of households in the vil- lage (excluding the respondent household) that have experienced upland reallocations, (iii) a dummy variable indicating whether the household believes that a reallocation
11
The literature on credit and technology adoption suggests that this variable is endogenous (i.e. correlated with unobserved factors of adoption such as entrepreneurial capacity). The test of endo- geneity conducted on this variable does not reject exogeneity, with various specifications and various measures of credit access. We therefore treat this variable as exogenous in the model.
12
is likely to occur before the end of the use right term, and (iv) the share of villagers (excluding the respondent household) expecting such a reallocation. These variables are only very weakly correlated13giving no cause for concern regarding multicollinear- ity. We include both household and village variables since decisions regarding land use are partly made at the village level. As a post-socialist country, Vietnam has a long tradition of collective decision making. This is particularly true in the rural ar- eas, especially regarding land use. Villages in the study area are mostly homogenous in ethnicity and even often constituted by households from the same clan. Conse- quently, the ties among villagers are usually strong, and the social life within a village is very intense, providing good information circulation within a village. Historically and nowadays, the ‘Vietnamese village’ is considered a strong entity in itself. Both the colonial administration and the communist party have later tried to instrumentalize the villages in order to impose political projects (see Bergeret (2003, p.30-33) for a detailed historical review of the “Vietnamese village myth”). After the decollectivisa- tion these structures have remained or even been reinforced. Sikor (2004) and Wirth
et al. (2004) show for Yen Chau how the organisation of villages has challenged the
implementation of the land law. At a larger scale, Kerkvliet (1995,2005) shows how ’everyday politics’ within Vietnamese villages have contributed to transform the na- tional policy at the time of the decollectivisation. Based on this evidence, we believe that it is plausible that villagers’ experiences and opinions significantly matter in a household’s decision regarding the adoption of agroforestry.