• Open Access

Evidence for the effects of environmental engagement and education on knowledge of wildlife laws in Madagascar


  • Editor
    Richard Cowling

Aidan Keane, Department of Life Sciences & Centre for Environmental Policy, Imperial College London, Silwood Park Campus, UK. Tel: 020 7594 2509; Fax: 020 7594 2339. E-mail: aidan.keane05@imperial.ac.uk


Rules are fundamental to the implementation of conservation policies, but cannot change behavior if they are not known or understood. Despite this, few studies have investigated knowledge of conservation rules or factors influencing it. Here, we quantify the effects of involvement with tourism and community-based natural resource management, education and demographic factors on local people's awareness of Madagascar's species protection laws. Knowledge of the laws was generally low. However, those who worked as tourist guides, hosted tourists, and were involved in local forest management committees were almost twice as likely to classify correctly a species as protected compared with individuals not exposed to conservation messages in this way. This year marks 50 years since Madagascar introduced its first species protection law. It is time to recognize that rules are necessary, but not sufficient, for species protection and to devote more attention to the communication, and enforcement, of conservation rules.


Rules and regulations are of central importance to conservation, underpinning a whole spectrum of approaches from community-based wildlife management and payments for ecosystem services to fishing quotas and protected areas (Keane et al. 2008). However, the existence of a rule does not guarantee that it will be respected (Rowcliffe et al. 2004). Various factors influence compliance with rules (Keane et al. 2008), but if they are not widely known, rules cannot change behavior. For example, studies of anglers in the United States (Page & Radomski 2006) and community-based natural resource management in Uganda (Nkonya et al. 2008) have shown that compliance is higher in groups with better awareness of the rules. It is therefore important for conservationists to understand which factors most strongly influence awareness of rules, but the topic has been neglected. Nkonya et al. (2008) found that, at the community level, awareness of locally enacted regulations protecting privately owned natural resources was lower among isolated groups, but was improved by the presence of environmental organizations. However, there has been no attempt to identify factors that improve awareness of rules at the individual level, where decisions about hunting and persecution are made.

Madagascar is one of the “hottest” biodiversity hotspots (Mittermeier et al. 2004). The island's diverse, highly endemic biota is threatened by habitat destruction (Green & Sussman 1990), overhunting (O’Brien et al. 2003; Golden 2009), persecution (Hawkins 2003), and collection for the pet trade (Andreone et al. 2005). Recent political difficulties have further exacerbated the situation (Barrett & Ratsimbazafy 2009). This year, it is 50 years since the first Malagasy wildlife law was passed to regulate hunting. Malagasy wildlife law (Law 60–126; Decree 2006-400) now divides species into three categories: protected (may not be hunted or killed), game (may be hunted only during specific periods and with a permit), and nuisance (not subject to any controls). There is considerable evidence that these laws are not well respected (Garcia & Goodman 2003; Goodman 2006; Jenkins & Racey 2008; Golden 2009), but there has been no investigation of how well they are known and understood.

This study quantifies influences on individuals’ awareness of species conservation laws among rural people in the eastern rainforest area of Madagascar. The involvement of local communities in tourism (Durbin & Ratrimoarisaona 1996) and, more recently, in the management of forests and their resources (Antona et al. 2004; Raik & Decker 2007) are key components of Madagascar's environmental policy, intended to improve conservation and resource management through a variety of mechanisms. We therefore ask whether involvement with these activities increases awareness of Madagascar's species protection laws. We also test whether having a higher level of education, often cited as a key determinant of environmental awareness (Howe 2009), improves knowledge of the law. Finally, we examine the effect of these factors on knowledge of the law among young adult males who have encountered the species in question, a key target group for rule enforcement likely to have the inclination and opportunity to hunt (Kümpel et al. 2010).


Study area

Our study was carried out in a rural area adjacent to the forest corridor that runs along Madagascar's eastern escarpment (Figure 1). Livelihoods in the region are based on small-scale farming with collection of forest products and hunting to supplement income and protein (Ferraro 2002; Jones et al. 2006). Between December 2007 and April 2008 interviews were conducted with 602 individuals from seven small administrative units known as fokontany within one commune.

Figure 1.

Map showing the location of each interview relative to the forest corridor and the commune centre. The inset shows the position of the Haute Matsiatra region, where our study was carried out, within Madagascar. In order to preserve the anonymity of respondents, the precise location is not given.

In four fokontany, management of natural resources within defined areas of forest has been devolved to community-based forest management organizations (Antona et al. 2004; Raik & Decker 2007). Individuals may choose whether or not to join the forest management organizations, although membership is subject to a fee. Some respondents were also members of the forest management committees.

The area receives a small but increasing number of tourists, with local environmental and development NGOs helping the commune to develop the area's eco- and ethno-tourism potential, specifically its primary rainforest and picturesque sacred mountain. Some respondents are engaged with the tourism industry, acting as tourist guides or hosting tourists to supplement their income from farming. The commune has no protected areas or major conservation interventions.

Data collection and analysis

Interviews were conducted in Malagasy, primarily by A. A. Ramarolahy, helped by a research assistant and a local guide. Participants were selected at random and questioned about 23 animal species found in the area (Supporting Information, Table S1). The interviews dealt solely with awareness of wildlife laws, and did not ask potentially sensitive questions about compliance. At the outset, the interviewer introduced himself as a student, described the purpose of the study, and assured them that their responses would be anonymous. First, participants were shown a photograph and asked to name it. If they were unable to identify the species correctly, they were given a hint and allowed to try again. The interviewees were then asked if they had ever seen the species and, finally, to indicate whether it was a protected, nuisance, or game species by placing the photograph into one of three piles. This procedure was repeated for each species, and the interviewer recorded whether each species was classified correctly or not. After the interview, the respondent's demographic characteristics were recorded (Supporting Information, Table S2).

Before analysis we discarded responses where a species was not correctly identified from the photograph and the hint (30.2% of the total) and those with missing data, leaving a sample of 8,059 responses from 542 individuals. The percentage of discarded responses was similar for each category, but varied between species and between different groups of respondents (Supporting Information, Table S3).

Statistical modeling

A series of multilevel logistic models were fitted to the data. The response was binary, indicating whether the respondent was able to place the species in the correct legal category. Eleven explanatory variables were grouped into six functional groups. The first was the species’ legal category. The remaining five described the respondents’ individual characteristics; their education (highest level of education attained), involvement with natural resource management or tourism activities (whether they pursue livelihood activities in addition to farming, such as guiding tourists; whether they belong to a household that hosts tourists; whether they belong to a forest management organization), demographic characteristics (age, sex), familiarity with the species (whether they have ever seen the species), and their location (the fokontany to which they belong, distance from the forest edge, distance from the commune center). We did not test specific hypotheses regarding the spatial location of respondents, but retained these predictors to control for the possibility of spatial patterns in the data.

Differences between species were modeled through the inclusion of a random effect because we were primarily interested in social rather than species-specific factors influencing awareness of conservation rules. This allowed us to quantify the variation between species without the need to estimate large numbers of parameters. Differences between species were examined informally by disaggregating the random effect. A second random effect, for individual, was included to account for the grouping structure of the data, since every respondent answered questions about each species.

A candidate set of models was chosen a priori. Each model included species category with varying combinations of the remaining groups of explanatory variables and their interactions with species category. The full set of 61 models was fitted in R 2.10.0 (R Development Core Team 2009) using the glmer function from the lme4 package, version 0.999375–32 (Bates & Maechler 2009). The Akaike information criterion (AIC) was used to rank the fitted models and construct a 99% confidence set. Model weightings were calculated based on this confidence set (Burnham & Anderson 2002).

Parameter estimates derived from these models are difficult to interpret directly because of the presence of interactions and nonlinearity. We therefore present average predictive comparisons, calculating the means and confidence intervals of responses simulated from the fitted models to evaluate their predictions at different values of one or more focal variables, holding all others constant (Gelman & Hill 2007). Both parameter uncertainty and model selection uncertainty were incorporated in these comparisons. Uncertainty in parameter estimates was incorporated by simulating every scenario 1,000 times, each time drawing parameter values at random from normal distributions whose means and standard deviations equaled the means and standard errors of the fitted models’ parameter estimates (Gelman & Hill 2007). Model selection uncertainty was incorporated by repeating this process for each of the models in the 99% confidence set, averaging their predictions weighted by their Akaike weights (Burnham & Anderson 2002).


The respondents’ ability to classify species into their legal categories was poor, with only 42.9% (n= 8059) correct responses in the raw data (cf. the expectation of 33.3% for uninformed guesses). However, there were substantial differences between the three categories. Nuisance species were correctly classified most often (63.8% of 1428 responses), followed by protected species (56.5% of 3137 responses), with game species rarely placed in the correct category (9.2% of 2494 responses).

Model selection resulted in a 99% confidence set of seven models (Table 1). The AIC-best model included all explanatory variables and their interactions with legal category, but only received a weighting of 0.31 reflecting a high degree of model selection uncertainty (Burnham & Anderson 2002). Other selected models dropped groups of variables representing education and familiarity with the species. Model-averaged predictions generated from the confidence set correspond well with the observed data (Figure 2).

Table 1.  Summary of the 99% confidence set of models selected based on AIC
 Model 1Model 2Model 3Model 4Model 5Model 6Model 7
  1. Note: The inclusion of different functional groups of predictor variables in each model is indicated by M (the main effects for these variables were included) and I (the interactions of these variables with species legal category were included). RE: species and RE: individual indicate the standard deviation of the random effects terms for species and individuals respectively. ΔAIC is the difference in AIC between the model in question and the AIC-best model (Model 1). w is the Akaike weight of the model.

DemographicM + IM + IM + IM + IM + IM + IM + I
EducationM + IM + IM + IMM  
NRM & tourismM + IM + IM + IM + IM + IM + IM + I
FamiliarityM + IM M + IMM + I 
Spatial locationM + IM + IM + IM + IM + IM + IM + I
RE: species1.
RE: individual0.540.550.550.540.550.550.56
Figure 2.

Model fit and variability attributable to random effects for species and individual respondents. The heavy black line indicates the predicted fit, averaged over the 99% confidence set of models. Black circles show the mean of the response variable binned into 1 unit intervals, with error bars of ±2 standard errors. The broken black lines and solid grey lines indicate the minimum and maximum values of the conditional modes of the random effects for individuals and species respectively, taken from the best fitting model, illustrating the range of variability.

Average predictive comparisons for combinations of species legal category and other predictor variables are presented in Figure 3. As in the raw data, the most important effect is that of species legal category, with nuisance species correctly categorized more often than protected species, while game species are almost always miscategorized. The effects of other predictors also interacted with legal category.

Figure 3.

Model-averaged average predictive comparisons illustrating the effect of each predictor variable on the probability of correctly categorising a species, and their interactions with the species' legal category. The dashed vertical line indicates the predicted overall mean response for the original dataset. The solid vertical line in each panel indicates the predicted mean response for the sample population if all of the species were protected, nuisance or game respectively. Heavy lines indicate approximate 67% confidence intervals, obtained by simulation. Light lines indicate approximate 95% confidence intervals. See supporting information Table S2 for descriptions of the variables.

The involvement of respondents with resource-management activities was associated with substantial improvements in categorizing protected species (Figure 3). Members of forest management associations were 21.1% more likely to categorize protected species correctly than nonmembers, but there was little difference between ordinary members and committee members. Predictions regarding the categorization of nuisance species were more variable, but there were indications that the members of forest management associations categorized nuisance species correctly less often than nonmembers. Hosting tourists had a lesser effect, but there was some indication that respondents from households that hosted tourists were slightly better at categorizing protected and game species, but worse at categorizing nuisance species.

Large differences were associated with occupation, with respondents holding an official position or acting as tourist guides being, respectively 25.0% and 36.4%, more likely to correctly categorize protected species than those who were just farmers. Guides were also more likely to classify game species correctly, but officials very rarely categorized game species correctly. Occupation produced no convincing effect on the categorization of nuisance species.

Respondents’ level of education was important for the categorization of protected species but had little effect for nuisance or game species. There was little difference between those with only primary education and those who had attended secondary education. However, respondents who were educated at a lycée were 24.1% more likely to categorize protected species correctly than those with only primary education.

We observed few clear differences associated with respondents’ demographic characteristics, but males were 22.0% more likely to categorize protected species correctly than females. There was also a slight improvement in the categorization of protected species with age, so that respondents aged 60 were 8.4% more likely to categorize protected species correctly than respondents aged 20. There was no clear effect of whether or not the respondent had ever encountered the species in question on their ability to categorize it correctly. There were substantial changes in levels of knowledge attributable to location. For example, respondents from fokontany A were 52.1% more likely than those from fokontany F to categorize protected species correctly.

From the perspective of a policy maker or conservation NGO, a key question is the extent to which education, or involvement with environmentally based activities such as tourism and local resource governance, affects awareness of laws among the individuals who most often hunt wildlife. We therefore used the model to predict how these factors change awareness in the group most likely to hunt, young (aged 25 years) male farmers with only primary education (Figure 4). Baseline levels of knowledge for this group are predicted to be much lower for protected species (47.2% categorized correctly) than for nuisance species (72.5% correct). However, guiding, membership of forest management organizations, and belonging to a household that hosts tourists all improved categorization, as does a lycée education. Individuals with all these characteristics were almost twice as likely to correctly classify protected species (89.1% correct) as those who did none of them.

Figure 4.

Average predictive comparisons illustrating the effect of conservation related activities and education on ability correctly to classify protected species amongst a target group of individuals likely to hunt wildlife. For the purposes of the scenario, this group was defined as young (aged 25 years) male farmers who have received only primary education. The solid vertical line indicates the baseline predicted mean response of the target population. Heavy lines indicate approximate 67% confidence intervals. Light lines indicate approximate 95% confidence intervals.

The fitted models suggest that approximately 21% of the remaining variation is attributable to variation between individuals, while 41% is between species. Of the protected species, lemurs were most often categorized correctly, followed by the birds and reptiles (Figure 5). The two carnivore species (Fossa fossana and Cryptoprocta ferox) and two protected insectivores (Setifer setosus and Limnogale mergulus) were least well categorized. In particular, the rare and cryptic aquatic tenrec (Limnogale mergulus) was very rarely categorized correctly.

Figure 5.

The conditional modes of the species random effect for protected species from the best fitting model (Model 1 in Table 1), indicating the differences in probabilities that the species were categorised correctly. Positive values indicate that a species was more likely to be correctly categorised. Heavy lines indicate an interval of ±1 SE and the lighter lines ±2 SE. Species are referred to by their Latin names. For common names, please refer to the supporting information, Table S1. The letters in brackets after each species correspond to their status on the IUCN Red List: LC = least concern, NT = near threatened, VU = vulnerable, EN = endangered (IUCN, 2009). Furcifer lateralis is not currently IUCN listed.


We found the level of knowledge about Madagascar's wildlife laws to be generally poor in our study area. One way to improve awareness of conservation rules is through dedicated education campaigns (e.g., Padua 1994), but these are expensive and can trade off with other conservation activities (Alder 1996). Consequently, it is important to know which factors predispose individuals to be better informed about conservation rules so that awareness-raising interventions can be effectively targeted.

For protected species, levels of awareness are substantially higher in better-educated individuals and those involved with tourism and community-based resource management. These findings are largely in agreement with those of previous studies that have examined the effects of ecotourism (e.g., Gadd 2005; Waylen et al. 2009), level of schooling (e.g., Howe 2009), and participation in community-based projects (e.g., Kideghesho et al. 2007) on awareness of and attitudes toward other aspects of conservation.

From a post hoc assessment, it is difficult to make robust inferences about the existence and nature of causal relationships (Morgan & Winship 2007). A statistical association between two variables (e.g., employment as a guide and awareness of conservation rules) might reflect a true causal process, but could also result if both are themselves correlated with another, unmeasured variable (e.g., interest in the forest). Furthermore, real causal linkages might not act solely in the hypothesized direction. For example, although we feel that an individual's awareness of the law is very unlikely to affect their probability of receiving employment as a guide or joining a forest management organization, rather than vice versa, the possibility cannot be entirely discounted from our data alone. Despite this limitation, we believe that an understanding of factors that correlate with awareness is useful. For example, these associations might be used to predict which subsections of a population are likely to be less aware of conservation rules, providing a starting point for targeting awareness-raising measures more efficiently.

Providing better education and creating tourism-based livelihood opportunities are common goals for development and conservation interventions, and improving awareness and understanding of wildlife laws is a useful byproduct of these activities. Currently, however, only a small subset of the population are guides or have been educated to lycée level (4 and 19 individuals, respectively). By contrast, the majority of the respondents (430 individuals) participated in forest management organizations. Previous studies have questioned whether community-based approaches to conservation can be effective (e.g., Agrawal & Gibson 1999). In Madagascar, the partial devolution of natural resource management to communities has shifted many responsibilities to the local level (Antona et al. 2004; Raik & Decker 2007), but forest management organizations have often received little support since their creation, and concerns have been raised that this could undermine their success (Hockley & Andriamarovololona 2007). However our results suggest that, irrespective of whether other benefits are realized, involvement with local forest management organizations helps to sensitize people to conservation laws.

Another striking finding is the very poor recognition of game species’ legal status. Although bushmeat has recently gained prominence as a conservation issue in Madagascar (e.g., Garcia & Goodman 2003; Goodman 2006; Golden 2009), the focus has been on protected species such as lemurs. The exploitation of game species has received very little attention, although it is likely to be wider in extent (Jenkins & Racey 2008). Our findings suggest that the laws regarding game species are currently too poorly known to stand any chance of influencing people's behavior.

In general, factors that improved respondents’ ability to categorize protected species also tended to improve categorization of game species, but due to smaller sample sizes the effects are less well estimated. By contrast, these same factors had little effect or even reduced the ability of respondents to categorize nuisance species. One interpretation is that increased exposure to conservation messages (through resource-management, tourism, and the like) biases individuals toward assuming, or reporting, that species are subject to legal protection.

Although our primary focus, here, was relating differences in awareness to respondents’ individual characteristics, we also observed species-related differences. Their causes are beyond the scope of this study but might reflect differing levels of agreement between national laws and local attitudes and beliefs. Although often viewed in isolation, national laws are part of a larger system of formal and informal rules recognized by local Malagasy, which incorporates traditional taboos or fady (Jones et al. 2008). In some cases, preexisting attitudes toward a species might correspond with its legal status. For example, the lemur Propithecus edwardsii, is legally protected and is also considered a taboo by many people in the area (Jones et al. 2008) so it might be expected to be categorized correctly more often than protected species that are not locally revered (such as the aquatic tenrec, known as “water rat”[voalavorano] in Magalasy). Awareness might also have been affected by taxon-specific conservation measures, such as the extensive efforts devoted to lemur conservation in many parts of Madagascar.

Changing people's behavior continues to be an important challenge for conservation and, in this respect, the creation and enforcement of rules is a vital part of the conservation toolbox. The processes that lead to behavioral change are, however, complex and remain under researched in conservation (Darnton 2008). Rules can only be effective if they are known and understood by the people whose behavior they are intended to regulate. Local awareness of rules is therefore a necessary condition for their success, and steps should be taken to raise awareness where it is lacking, but awareness is only the first step toward achieving compliance. Further work is needed to understand how awareness of rules can be raised efficiently, and the extent to which changes in awareness translate into changes in compliance. Reducing rule-breaking in conservation would also benefit from an improved understanding of factors that influence individuals’ decisions about whether to obey known rules, including enforcement measures and other incentive-based interventions, in order to build an evidence base for the creation of robust, successful, and scientifically informed policies to promote behavioral change.


We are grateful to the Leverhulme Trust, Economic and Social Research Council and a Royal Society Wolfson Research Merit Award for funding this research. We would also like to thank Madagascar's Ministry of Water and Forests and Bruno Ramamonjisoa of the École Supérieure des Sciences Agronomiques, Université d’Antananarivo for their help and assistance, and two anonymous reviewers for constructive suggestions that helped us to improve the manuscript.