• Open Access

Predicting willingness-to-sell and its utility for assessing conservation opportunity for expanding protected area networks


  • Editor 
    Belinda Reyers

Angela M. Guerrero, The University of Queensland, School of Biological Sciences, Brisbane, Queensland 4072, Australia. Tel: +61 (7) 33652494; fax: +61 (7) 33651655. E-mail: a.guerrero@uqconnect.edu.au


Translating maps of priority areas for conservation into activities which actually secure these places is a complex process, overwhelmingly influenced by human and social factors. Spatial conservation prioritizations often assume that land is available for acquisition. Using data gathered through interviews, we predicted land managers willingness-to-sell their land. We included this information in spatial prioritization analyses that aimed to identify areas that achieve conservation objectives while maximizing the probability that land will be available for acquisition. As anticipated, our solutions were more likely of being available for acquisition compared to solutions that did not include the willingness-to-sell data. Our results illustrate the trade-off between designing so-called “optimal” networks of protected areas, and the realities of translating these maps into on-ground action. We demonstrate how an important aspect of the social context in which conservation is embedded and defines conservation opportunities can be explicitly accounted for in the development of conservation plans.


The gazettal of formally protected areas is critical for ensuring the persistence of biological diversity (McNeely 1994; Redford & Richter 1999). Prioritizing where, and increasingly, when and how, conservation investments should occur is an important part of the conservation planning process (Knight et al. 2006; Wilson et al. 2007; Pressey & Bottrill 2009). One of the challenges faced by conservation planners when prioritizing areas for acquisition is accounting for the complex social, economic, and institutional environment in which conservation is embedded (Balmford & Cowling 2006; Knight et al. 2006). A failure to do so can result in recommendations that are not able to be implemented, leading to a reduced delivery of conservation outcomes. To avoid this, spatial prioritization analyses should be based not only on an assessment of conservation values, but also an assessment of the opportunities for, and constraints upon, implementation, which are often defined by the characteristics and needs of stakeholders (Knight & Cowling 2007).

Increasingly, conservation scientists are expanding the conceptual basis of conservation planning beyond the natural sciences and efforts to improve the effectiveness of spatial prioritizations have manifest in several ways (Balmford & Cowling 2006; Pannell et al. 2006). The importance of the social sciences within the process of conservation planning has been recognized (Balmford & Cowling 2006; Polasky 2008). Specifically, the importance of fostering stakeholder participation (DeCaro & Stokes 2008), social capital (Pretty & Smith 2003; Cramb 2006), and social learning institutions (Knight et al. 2006) has been identified as essential for effective conservation planning initiatives. This has lead to the development of operational models for conservation planning that incorporate phases dedicated to stakeholder collaboration, social assessments, and the development of implementation strategies (Knight et al. 2006; Cowling & Wilhelm-Rechmann 2007; Pressey & Botrill 2009). There have also been attempts to include social and economic objectives explicitly into spatial prioritization analyses (Stewart & Possingham 2005; McBride et al. 2007), which has lead to a conceptual shift from assessing conservation priority to conservation opportunity (Noss et al. 2002; Knight & Cowling 2007). This has been complemented with a furthering of our understanding of the human and social factors that drive conservation success (Tenge et al. 2004; Pannell et al. 2006; Dutton et al. 2008). However, no examples exist where knowledge of these factors has been used to predict the feasibility of acquiring land for conservation and inform the development of a conservation plan.

We use predictions of land manager's willingness-to-sell their land to a conservation organization as a surrogate measure of the probability that the land might be acquired for conservation. We show how predictions of willingness-to-sell can be used in a spatial prioritization analysis to generate conservation plans that have a higher probability of being available for acquisition. We test whether a land manager's willingness-to-sell can be predicted from census data as opposed to data available only through face-to-face interviews. We compared achievement of conservation targets of conservation plans identified by spatial prioritization analyses using predicted land manager willingness-to-sell data to analyses that did not explicitly account for such information.


Study area

The study area comprises a portion of the Fish-Kowie corridor, one of seven conservation corridors identified throughout the Subtropical Thicket Biome for ensuring the persistence of globally important nature (Rouget et al. 2006). It forms part of the south-western portion of the Maputaland-Pondoland-Albany hotspot, an area with exceptional levels of plant endemism (Steenkamp et al. 2004). Specifically, the study area comprises 301 land parcels owned by 48 land managers totaling 146,660 hectares (Figure 1), and located within the Makana Municipality of the Eastern Cape Province of South Africa. The dominant land uses are eco-tourism, and game and livestock farming. While current rates of vegetation clearing are low compared to historical rates across the study area (Knight et al. 2010) major pressures upon species and habitats include overgrazing by livestock, harvesting of plants for traditional medicinal uses, and invasive alien plant species encroachment (Makana Municipality 2008).

Figure 1.

The Makana Municipality study area in the Eastern Cape Province of South Africa.

Eastern Cape Parks Board's land acquisition policy is to extend protected areas in the Province by 10,000 hectares per annum (Eastern Cape Parks 2009); expanding the Great Fish Nature Reserve is one of the organization's priorities. Given the location of the Great Fish River Reserve within the study area, a strategic opportunity exists for expanding the Eastern Cape's protected area network through the acquisition of private land.

Data collection and analysis

Our conservation objective was to increase the representativeness of the reserve network by improving the protection of 19 different vegetation types (Vlok et al. 2003). Cadastral data from the Surveyor-General's Office were used as planning units, as they are the legal unit of land purchase and transfer. The average cost of land acquisition in the study area is estimated to be US$608 per hectare, based on 2006 data of the South African Property Transfer Guide and 2001 data from the Surveyor General's Office (Knight et al. 2010).

Semi-structured, face-to-face interviews were conducted with 48 land managers in the study area to collect data on a range of demographic, attitudinal, and behavioral characteristics that were hypothesized to define opportunities for private land conservation (see Knight et al. 2010, for a detailed explanation of the sampling procedure). The collected data included a question statement to assess land managers’ willingness to sell their land for conservation purposes. The question statement used a 1 to 5 rating scale from “strongly disagree” to “strongly agree,” which were reclassified into a high willingness-to-sell and a low willingness-to-sell. Using the statistical package SPSS 15.0 (2006) we analyzed the survey responses to identify potential predictors of a land manager's willingness-to-sell their land to a nature conservation organization.

Probabilistic models of willingness-to-sell

We converted the willingness-to-sell responses to probabilistic data using binary logistic regression analysis. We built three probabilistic models each based on different types of survey data, specifically: (1) data that could only be sourced from interviews with individual land managers, (2) data that could be derived from census or other more general sources, and (3) a combination of survey and census data. Candidate models were constructed using the identified potential predictors of willingness-to-sell. All models were evaluated for their goodness-of-fit via Pearson chi-square and Hosmer and Lemeshow statistics, and receiver operating characteristic curves (ROC). They were compared based on indices of parsimony measured via the Akaike's information criterion and their predictive power measured via pseudo R-square measures. The best-performing models according to these metrics were identified.

Spatial prioritization analysis

To identify priority areas for conservation investment we used the spatial prioritization software Marxan (Ball et al. 2009). Marxan uses a mathematical objective function to identify near-optimal solutions that achieve explicitly stated conservation targets and seeks to minimize an objective function. To account for the probabilities reflecting the willingness of the land managers to sell their land we used a modified objective function that incorporates a probabilistic component (Equation (1)):


Where the planning units i are each of the 301 land parcels in the study area, and the vegetation types j are the 19 identified by Vlok et al. (2003). The purchase cost of each planning unit is the estimated average sale price, and the “species penalty factor” (SPF) is the penalty cost for not meeting the conservation targets for each vegetation type. We considered two conservation targets—10% and 30% of the extent of each of the 19 different vegetation types found in the study area. The 10% target is a generally applied (McNeely 1994), though widely criticized standard (Soulé & Sanjayan 1998; Carwardine et al. 2008) and the 30% target represents a more recent international recommendation (IUCN 2003). Each planning unit was weighted by a probability that it will not be available for sale to a conservation organization, derived from the three probabilistic models. The Boundary Length Modifier (BLM)—controlling the compactness of the solutions—was set to zero as simultaneously attempting to address connectivity issues would confound our ability to isolate the effects of land managers willingness-to-sell.

The simulated annealing algorithm employed in Marxan attempts to minimize the value of the objective function—Equation (1). The “best solution” is the solution generated with the lowest score, which is a combination of the purchase cost, penalty cost, and probability.

We first ran a base analysis without consideration of the probability component. The objective of this analysis was to find the set of planning units that minimize the overall purchase cost subject to all vegetation types meeting their targets. To reflect this constraint we employed an SPF of 3, which ensured that all vegetation types met their targets, for both the 10% and 30% target scenarios.

The budget required to meet the vegetation targets in the first analysis was employed as the constraint in the second analysis. The objective of the second analysis was to maximize the number of targets achieved and the probability of planning units being available for acquisition, given the fixed budget. For each analysis, we explored two target levels (10% and 30%) and three different data sets using the findings of the three probabilistic models.


Social composition of the study area

Surveyed land managers were mostly male (90%) and married (92%) with an average of two children. On average, they had lived on their property for 23 years and had owned it for 14 years. Most were experienced land managers, with three-quarters having farmed for more than 5 years, and half having farmed for more than 20 years. Most of the land managers generated their income from on-farm activities (67%), particularly grazing of small stock (e.g., sheep, goats).

Predicting land managers’ willingness-to-sell

The statistical analysis revealed several factors that had a relationship with a land manager's willingness-to-sell their land to a nature conservation organization. These factors included both demographic and attitudinal characteristics (Table 1). The results suggest that the length of time a land manager has owned the property, and how long they have been farming influences this willingness. Human factors such as a land manager's conservation knowledge and conservation behavior, and their willingness to be involved in conservation initiatives were also identified as potential predictors. A number of social capital factors that related to a land manager's local and broader networks, and their willingness to collaborate, were also found to be associated with their willingness-to-sell. Lastly, a land manager's potential for burnout also showed an association to their willingness-to-sell to a nature conservation organization.

Table 1.  Demographic, human capital, and social capital factors that are related to, and are potential indicators for, a land manager's willingness to preferentially sell land to a conservation organization
Human/social factorSignificance testing result (probability of type I error)a
  1. aThe relationship between each factor and willingness-to-sell was found to be positive (for nominal variables, e.g., Primary land use or Cultural group, the relationship with willingness-to-sell is with the secondary category). The relationship between the categorical variables and willingness-to-sell was explored using contingency tables and the Pearson chi-square test. Analysis of variance (ANOVA) test was used to explore the relationship between the continuous variables and willingness-to-sell. Multicollinearity was evaluated by calculating partial correlations and variance inflation factors. The results of these analyses found no multicollinearity between potential predictors.

 Years of ownership0.15
 Years of farming0.12
 Primary land use (Grazing or Game)0.09
 Number of generations property has been owned by family0.02
 Land manager's cultural group (White English or White Afrikaans)0.05
Human capital
 Conservation knowledge—awareness of permit requirements to clear land0.09
 Conservation behavior—Running ecotourism activities in farm0.03
 Conservation behavior—undertakes soil conservation measures0.19
 Conservation behavior—undertakes conservation activities for plants0.19
 Willing to be involved in a binding conservation agreement0.08
 Willing to be involved in a common property resource management agreement0.00
 Willing to be involved in a voluntary conservation agreement0.15
 Willing to be involved in a Landcare group0.06
 Interest in receiving signage for voluntary conservation agreement membership as an incentive for nature conservation0.09
 Would consider reducing production activities even if offered no incentives0.11
 Would consider reducing production activities if offered appropriate incentives0.06
 Entrepreneurship—ability to identify new business opportunities0.07
 Burn-out potential—“impersonal” treatment of some staff0.02
 Burn-out potential—land manager “working too hard”0.05
Social capital
 Local networks—Membership of sport clubs (cricket club)0.07
 Local networks—Membership of sport club (hunting club)0.08
 Broader networks—attendance at local municipality meetings0.18
 Broader networks—member of E.C. Game Man. Association0.09
 Membership of local social organizations—children school group0.16
 Membership of local social organizations—Agricultural shows0.10
 Willingness to collaborate—with police service0.19
 Willingness to collaborate—with Farmers Association0.05

Using the potential predictors, several probabilistic models to estimate the likelihood of a land manager selling their land to a conservation organization were developed. An evaluation of each model's performance led to the selection of three probabilistic models (A–C). Model A was chosen based on its overall performance across a variety of statistical metrics and included a mix of demographic and human and social factors. Models B and C were chosen to reflect two different types of predictor variables—survey-based and census-based data. Both were also among the best performing models. The predictor variables used in each model are shown in Table 2.

Table 2.  Binary logistic regression models for predicting the likelihood that land might be available for conservation purchase
ModelPredictor variables—demographic, human, and social factorsStandardized CoefficientSignificance testing result
  1. a Sample size = 45; Chi-square Hosmer & Hemeshow = 0.3; Prediction rate of 87%; Pseudo R2 (Cox & Snell 1989; Nagelkerke 1991) = 0.47, 0.63; ROC 0.94.

  2. bSample Size = 46; Chi-square Hosmer & Hemeshow = 0.7; Prediction rate of 83%; Pseudo R2 (Cox & Snell 1989; Nagelkerke 1991) = 0.43; 0.58; ROC 0.95.

  3. c Sample size = 45, Chi-square Hosmer & Hemeshow = 0.3; Prediction rate of 82%; Pseudo R2 (Cox & Snell 1989; Nagelkerke 1991) = 0.40, 0.54; ROC 0.87.

  4. Note: sample sizes are lower than 48 due to missing data (i.e., some land managers left some questions unanswered).

Model A Best performing modelaYears of living on the property0.940.01
Member of Farmers Association31.80.01
Burnout potential—impersonal treatment of staff0.060.03
Land manager's cultural group (White English or White Afrikaans)0.320.33
Model B Survey-basedbBurnout potential—impersonal treatment of staff0.020.01
Burnout potential—land manager “working too hard”8.100.10
Willingness to participate in conservation initiatives (index)12.990.01
Model C Census-basedcMember of Hunting Club9.010.06
Member of EC Game Management Association1.930.50
Number generations property has been owned by the family0.380.02
Member of cricket club2.330.54
Land manager's cultural group (White English or White Afrikaans)0.460.50

Spatial prioritization analysis

Using predictions of land managers’ willingness-to-sell we identified solutions comprising planning units that were on average 1.4 to 1.6 times more likely to be available for acquisition (Table 3). The solutions generated by the three probabilistic models were similar (Table 3). Including predictions of land managers’ willingness-to-sell into the analysis (while controlling for the overall cost of the solution) resulted in solutions with lower target achievement. For the 10% target scenario an average of 14 of the 19 vegetation types met their targets, while only 11 of the 19 vegetation types met their targets for the 30% target scenario.

Table 3.  Spatial prioritization analysis results
  Acquisition Cost (US$‘000s)Targets achievedaAvailability for acquisitionb (average probability across all planning units, overall probability)
Model AModel BModel C
  1. aTarget achievement refers to the representation targets for each vegetation type.

  2. bAvailability for acquisition reflects the probability that planning units in the resulting solutions will be available for acquisition. The probabilities are generated by the use of the Binary logistic regression models for predicting the likelihood that land managers might be willing to sell their land to a conservation organization (Models A, B, and C).

  3. c The solutions that meet all targets at a minimum cost.

  4. d Given a fixed budget, the solutions that maximize the number of targets achieved and the probability of planning units being available for acquisition.

Conservation Target = 10%
Base Analysisc—with no 9,0671963%, 2.4 × 10–2357%, 2.5 × 10–3263%, 5.8 × 10–21
 consideration of
 “willingness-to-sell” probabilities
Scenario Analysisd—withModel A8,9051688%, 1.6 × 10–3
 Acquisition ofModel B9,0661388%, 1.9 × 10–5
 “willingness-to-sell” probabilitiesModel C9,02312  86%, 8.5 × 10–4
Conservation Target = 30%
Base Analysisc—with no 26,4631963%, 2.8 × 10–4157%, 8.2 × 10–6061%, 8.2 × 10–45
 consideration of
 “willingness-to-sell” probabilities
Scenario Analysisd—withModel A25,8691286%, 3.2 × 10–9
 consideration ofModel B26,0741090%, 1.8 × 10–7
 “willingness-to-sell” probabilitiesModel C26,04310  85%, 7.7 × 10–8


Our study shows how predictions of willingness-to-sell can be estimated and used in spatial prioritization analyses to generate solutions that have a higher likelihood of being available for acquisition. Spatial prioritization analyses that assume all land is equally available for acquisition will not only likely result in a plan that cannot be implemented, but that also wastes the limited resources available for conservation through necessitating that analyses be repeated and implementation costs (e.g., for negotiation) be unnecessarily wasted.

In our example, conservation targets for some conservation features were compromised when the willingness-to-sell data were included, and therefore not only might it be impossible to implement an optimal solution, but also that finding a feasible solution will likely entail trade-offs. In other spatial prioritization problems, an abundant choice of planning units to meet conservation targets might allow the targets to be compromised to a lesser extent (Game et al. 2008). When there is a trade-off, the appropriate sacrifice (e.g., additional expenditure, or reduced target achievement) will require negotiation between stakeholders.

Our study demonstrates how a diverse set of social and human factors can be effective predictors of the likelihood that a land manager is willing to sell their land to a conservation organization. In our study, a land managers membership of specific organizations is a good indicator of the willingness-to-sell their land. Another demographic variable found to be a reliable predictor of willingness-to-sell is the time land has been owned by a family. There is evidence that this variable has a positive relationship with conservation attitude in certain regions (Winter et al. 2005), but not in others (Wilson 1996), demonstrating that predictors of land manager willingness-to-sell, and probably other dimensions of conservation opportunity, may differ between regions. A land manager's willingness-to-participate in private land conservation programs was a good predictor of their willingness-to-sell their land to a conservation organization. More generally, the relationship between attitudes toward conservation and effective conservation action has been previously explored (Astrat et al. 2004; Winter et al. 2005; Barr & Gilg 2007). Of the factors we found to have a relationship with a land manager's willingness-to-sell their land, burnout has been the least documented. Burnout, defined as “a syndrome of emotional exhaustion, depersonalization, and reduced personal accomplishment” (Maslach et al. 1996), has been explored in voluntary natural resource management programs on private land in Australia (Byron et al. 2001) and has been hypothesized to define, in part, conservation opportunity (Knight et al. 2010). Our study confirms that indicators of burnout can be used to predict opportunities for conservation, specifically willingness-to-sell.

Data gathered through face-to-face interviews will provide the most accurate data on land manager willingness-to-sell, and more generally on conservation opportunity, as people display idiosyncratic attitudes and behaviors, both within, and between, regions. However, gathering, analyzing, and incorporating human and social data into a spatial prioritization can be time-consuming and costly, particularly at regional scales, meaning that prohibitive numbers of people must be interviewed to obtain a representative sample. Approaches that predict willingness-to-sell based on existing or remotely collected surrogates of survey data are therefore highly desirable. Our Model C, based on data able to be collated through census or organizational records, performed as well as the model based on interview data. This result potentially provides an approach for cost-effectively scaling-up local-scale data for regional-scale application. However, face-to-face interviews serve not only to provide data, but also serve as a mechanism for engaging and informing stakeholders in a conservation planning process. This benefit should be pursued, if not through surveys, through other genuine mechanisms of stakeholder collaboration.

This research explicitly aims to recognize that conservation is a social process (Knight et al. 2006) and is dependent on people and the choices they make (Cowling & Pressey 2003). However, generalizations about human and social factors influencing the likely implementation of conservation actions should be made with caution (Potter & Lobley 1992; Winter et al. 2005). An analysis of factors that define willingness-to-sell, or conservation opportunity more generally, is likely specific to the context of the planning region under investigation (Knight & Cowling 2007). Also, people's preferences and choices are dynamic and can change with time given the influence of a diversity of external factors. Therefore any generalizations should be field validated.

This research could be expanded in several ways. First, our analysis could be expanded to examine alternative measures of conservation opportunity, not just the availability of land for protection. While some land managers might be unwilling to sell their land to conservation organizations (e.g., they plan to transfer the property onto the next generation) they might be willing to engage in private land conservation activities (see Knight et al. 2010). The human and social factors that predict the availability of land for acquisition might differ from the factors that predict stakeholder participation in other activities. Second, a better understanding of the relationship between a land manager's willingness to engage specific conservation instruments and institutions and their actual behavior could provide more reliable models for targeting conservation activities (Barr & Gilg 2007). A predictive model, similar to the one presented in this study, could be developed from research into actual behavior.

To ensure that conservation plans are translated into effective conservation action, there is a need to explicitly include the social and human factors that drive conservation opportunity (Knight & Cowling 2007). This requires spatial prioritization analyses that not only focus on meeting conservation targets but that explicitly account for the social process in which conservation is embedded (Cowling & Pressey 2003; Balmford & Cowling 2006; Knight et al. 2009). This research illustrates the importance of accounting for the socioeconomic context in spatial conservation prioritizations, specifically, the likelihood of land being available for acquisition.


We sincerely thank the land managers who participated in the survey. Comments from A. Ainslie, L. Pasquini, and S. Shackleton improved the interview approach. G. McGregor provided transport and cartographic support. M. Watts provided analytical advice. The Global Environment Facility, The World Bank, the Department of Botany at Nelson Mandela Metropolitan University, the Department of Environmental Science at Rhodes University, and The Australian Research Council provided funding and support.