Understanding migration to protected area buffer zones in Costa Rica utilizing cultural consensus analysis

. Human migration to the world’s protected areas’ (PA) buffer zones is widely seen as a significant threat to conserving biodiversity. Research since 2005 has demonstrated some evidence for global migration trends but also highlighted the simultaneous need to understand the local, contextual factors that drive migration around individual PAs. Investigation into human migration patterns to these buffer zones has frequently relied on methods that do not accurately capture the calculus used by migrants in their decisions. The research presented here uses a mixed-methods, cognitive anthropological approach to assess the motivations of Costa Rican migrants to the buffer zones of three national parks. Employing cultural consensus analysis methodology in combination with a demographic analysis based on the Costa Rican census, this study was able to develop important insights into Costa Rican migrant motivations. Importantly, the research finds that there is not a single cultural model among the migrants surveyed regarding conditions driving their decisions. However, data collected indicate significant trends in migrants’ evaluation of critical variables driving decisions, how they relate to one another, and their significance to these migrants. Thus, migrant assessments of the conditions of these variables in both previous and current communities reveal a more complex, contextual picture. This work demonstrates the potential of cognitive anthropological methods to help unpack migrant decision making and help conservation managers understand the factors that drive migration to surrounding communities. The analysis provides further evidence supporting calls for methods that help managers and communities understand the particularities of migration behavior in PA contexts.


INTRODUCTION
The relationship between population growth in protected area (PA) buffer zones and its potential impacts upon biodiversity is a concern for conservation biologists, practitioners, policy makers, and social scientists (Ypsilantis 1992, Harmon and Brechin 1994, Scholte 2003, Hoffman 2017, Oglethorpe et al. 2007). The negative effects of habitat fragmentation on biodiversity resulting from population-induced pressure on resources within and outside PAs is a primary concern (Hansen and Rotella 2002, DeFries et al. 2005, Hansen and DeFries 2007. Fragmentation in human-dominated landscapes outside PAs can increasingly isolate plant and animal populations with concomitant impacts on genetic flow, decrease species resilience to changing environments and the effects of climate change, and lower species richness both outside PAs and inside PA borders (Sherbinin and Freudenberger 1998, Hansen and DeFries 2007, Estes et al. 2012, Bamford et al. 2014. Another issue derived from increasing human populations on the borders of PAs is the threat of direct consumption and trade in plants, trees, and wildlife (Bamford et al. 2014). It is, therefore, critically important for biodiversity conservation and conservation practitioners to understand the processes that drive population growth in PA buffer zones.
As with all human population growth there are two possible explanations of PA buffer zone population increases: localized birth rate increases or the arrival of new populations via migration. This paper engages the latter causal mechanism with the specific aim of investigating the ways that PAs and PA policies are linked, or not, to migrant decision making. George Wittemyer and colleagues' seminal paper in Science (2008) shows that excessive population growth measured on PA edges is rooted in migration, not natural population growth. Their study, based on 306 PAs in Africa and Latin America, showed population growth in PA buffer zones exceeding that of comparable rural areas (Wittemyer et al. 2008). Further, Wittemyer et al. (2008) hypothesized that this migration was the result of the socioeconomic benefits created by the dominant conservation praxis of the late 20th century, specifically conservation and development projects. They suggested that migration-based growth is encouraged by PAs because conservation and development policies provide economic incentives (e.g., employment opportunities), ecosystem services (e.g., natural resources), and infrastructural benefits (e.g., roads) not present in similar rural locations, which "pull" migrants to PA buffer zones.
The Wittemyer et al. (2008) article is widely cited, often uncritically (Hoffman 2017). An analysis conducted in June 2020 showed over 350 citations according to Scopus and over 620 citations according to Google Scholar. That said, the findings and hypotheses from the article have been critiqued from several angles. First, some explicitly question the methods and analysis upon which their hypotheses are based (Joppa et al. 2008, 2009, Joppa 2012. Others caution that the hypothesized connection between PAs, conservation and development policies, and human migration to buffer zones is not fully supported when specific conservation contexts are examined (Davis 2011, Fay 2011, Hoffman 2011, Estes et al. 2012, Zommers and MacDonald 2012, Bamford et al. 2014, Salerno et al. 2014, Gupta 2015, Hartter et al. 2015, Ament and Cumming 2016, Brambilla and Ronchi 2016, Cripps and Gardner 2016. To their credit, Wittemyer et al. (2008) suggested that the hypothetical explanations based on their meta-analysis needed to Since its publication, some work has explicitly explored the veracity of Wittemyer et al.'s hypotheses on the relationship between PAs and migration (Hoffman 2011, Levang et al. 2012, Guerbois et al. 2013, Baird 2014, Bamford et al. 2014, Salerno et al. 2014, Gupta 2015, Hartter et al. 2015. Most analyses are based upon a combination of demographic, geospatial, and interview / survey methods to determine the following: (a) demographic changes and migration patterns; (b) locations of demographic and ecological change; and (c) attitudes and/or motivations of migrants in communities within the buffer zones of PAs. The social science methodologies are dominated by focus group, survey, and oral history approaches, and only a few studies exclusively focus on understanding migrant perspectives. At regional or PA-level scales, these authors find some evidence to support Wittemyer et al.'s (2008) "attraction model," but state that refining the scale of analysis illuminates variation based upon demographics, environmental factors, and political-economic history. Yet, none of these studies explicitly utilize cognitive approaches to determine whether migrants to PA buffer zones think similarly about the relationship between migration and specific, contextual factors.
Thus far, the research on PAs, population growth, and migration has not demonstrated a singular pattern, and testing Wittemyer et al.'s hypotheses in relation to cognitive data explicitly derived from migrant perspectives and motivations was not yet attempted. In order to fill this gap in knowledge and test the hypothesized drivers or PA-buffer zone migration, we examine Costa Rican migrants' motivations for relocating to the 10-km buffer zones of three Costa Rican National Parks: Carara, Arenal, and Barra Honda. Ten km buffer zones were intentionally chosen in order to replicate the limits employed by Wittemyer et al. (2008). We combine cognitive anthropological methods, particularly cultural consensus analysis (CCA), to examine the relationship between PAs, human migration, and population growth in PA buffer zones. CCA tests the shared knowledge of a societal sub-group regarding a particular cognitive realm or "domain," in this case Costa Rican migrants and their motivations for migrating to PA buffer zones. It is important to note that this study specifically focused on the internal migration of Costa Ricans and not that of foreigners. By applying these methods to three different national park contexts in Costa Rica, we test whether a general cultural model of migration can be built and whether Costa Rican buffer zone migrants' motivations reflect the hypothesis proposed by Wittemyer et al. (2008).

Rationale for parks selected
We intentionally selected Carara, Arenal, and Barra Honda national parks (see Fig. 1) out of the 26 national parks and the 160 PAs in Costa Rica for several reasons. First, it was important to maintain consistency by only investigating national parks. Costa Rica has various categories of PA ranging from those that do not allow resource use (i.e., national parks) to those that integrate resource exploitation into its mandate (i.e., protected zones). Standardizing the park category ensured consistency and removed spurious correlations related to the potential for legal resource use within PA borders. Second, all three parks were included in the data set that formed the basis for Wittemyer et al.'s (2008) original analysis. Third, this approach differs from Wittemyer's indiscriminate use of PAs that failed to create a consistent and comparable sample for which they were already critiqued (see Hoffman 2011). Finally, earlier published research showed very divergent characteristics leading to different migration patterns in these parks (Hoffman 2011).

Fig. 1.
Map of the study area. The study was set in the 10-km buffer zones (concentric lines) of three Costa Rican national parks (from left to right): Barra Honda National Park, Arenal Volcano National Park, and Carara National Park. The peninsula in the image is the Nicoya Peninsula found in the northwest of the country.
Carara is characterized by a buffer zone with explosive population growth due to internal migration, but this is largely unrelated to the park in itself and could be characterized as "incidental." Carara is located on the central Pacific coast and very close to several growing beach communities such as Jacó and Playa Herradura. Arenal was seen as the best fit for Wittemyer et al.'s "pull" model and potentially provided insight into migration to buffer zones that fit an idealized conservation and development paradigm. Arenal is famous for its picturesque namesake volcano and the ecosystem services it provides, such as hot springs, along with a diverse set of conservation and tourism development initiatives within a 40-km radius. Thus, the Arenal region supports a robust ecotourism-based regional economy in addition to strong agricultural production. Last, Barra Honda had virtually no population growth and very little internal migration to its buffer zone that enabled it to function as a "control" in which the link between development, migration, and the park is largely non-existent. Barra Honda offers its unique cave system for exploration, but there is little https://www.ecologyandsociety.org/vol27/iss4/art16/ other tourist attraction to the area, and the surrounding communities are characterized by a rural economy in decline for decades. It is precisely because of these marked variations in context that they were chosen.

Cultural consensus analysis
Cultural consensus analysis (CCA) allows social scientists to test whether a group of people has a set of knowledge or perceptions upon which there is agreement / consensus (Miller et al. 2004, Bernard 2017. CCA is a statistical method that grew out of cognitive anthropology and cultural consensus theory (CCT), which views culture as sets of knowledge that are shared by members of a society. From this perspective, it is the sharing of and consensus about this knowledge that forms the basis for what is commonly referred to as culture (d' Andrade 1984, Dressler et al. 2005, Copeland 2011). CCT assumes that groups share a single cultural model and that social scientists can determine the degree to which there is consensus in a group about a certain cultural domain via mathematical testing (Romney et al. 1986, 1987, Borgatti 1994. A domain is an organized set of words, concepts, sentences, all on the same level of contrast, that jointly refer to a singular conceptual sphere (Weller and Romney 1988). Thus, CCA allows researchers to probe variation within groups and to decipher cultural "truths" and the degree to which these truths are shared in a statistically meaningful, reliable, and replicable way (Strong and White 2020). By acknowledging informants' given answers as being probabilistic rather than inherently true, CCA searches for cultural truths not in individual responses but in the degree of sharing of these responses (Strong and White 2020).
Statistically, CCA works from a correlation matrix of respondents to estimate a linear function summarizing the similarities in their ratings (Dressler et al. 2018). There are two general approaches to operationalizing CCT theory in the form of CCA, the formal and informal model. This work uses the informal model, which is essentially a factor analysis (principal components analysis) of people (Weller 2007). In the informal model, each participant is given a competence score that tells the researchers how well each individual's responses correspond with those of the group. The informal model is, therefore, a set of statistical procedures that estimates both the answers to questions and respondent accuracy for answering those questions, which is calculated as a competence score (Weller 2007). A competence score is a measure of the degree to which participants' individual knowledge fits to that of the overall model developed from the entire group (Romney et al. 1986, Dressler et al. 2005, Copeland 2011). The correlations of individuals to the linear function provides their cultural competence, or the degree to which their understanding of the domain corresponds to the aggregated knowledge (Dressler et al. 2018). Individuals' responses can then be combined in a weighted average (weighted by each respondent's competence), which is referred to as the culturally correct answer key that is, in essence, how a culturally competent member of that society would answer those questions (Dressler et al. 2018). In sum, CCA facilitates the determination of the existence of a cohesive cultural domain, the cultural competence of each respondent, and the culturally "correct" answer key to the survey (Strong and White 2020).
Whether the data demonstrates a shared cultural model is dependent on several results from the analysis, which are processed using software packages. In this study, we utilized Anthropac (4.98) for our CCA analyses. CCA literature consistently states that the standard thresholds and minimum requirements to indicate strong consensus are the following: (1) a 3:1 ratio of the first factor eigenvalue to the second factor eigenvalue; (2) an average competence score of 0.6 or greater; (3) no negative competence scores; (4) at least 20 questions or "factors"; and (5) a minimum number of 30 participants, although this last point can be surmounted by a higher average competence score indicating model validity (Weller 2007, Bernard 2017. There is broad agreement regarding these thresholds, but others note that they are merely "rules of thumb" and that their strict use for indicating consensus within a group can be challenged mathematically (Purzycki and Jamieson-Lane 2017). As long as the other requirements are met, an average competence score of 0.5 already indicates consensus and that anything above 0.66 indicates strong consensus among the group (Weller 2007). Last, it is "by custom" that we use the first eigenvalue being three times larger than the second to determine that shared culture is driving the answers (Weller 2007).
Ultimately, CCA provides researchers with the ability to systematically verify whether there is a shared set of knowledge, and how well that knowledge is shared, amongst a group (Gatewood 2012). It was precisely CCA's capability to detect shared knowledge that drove our decision to employ it in our analysis of human migration to the buffer zones of Costa Rican national parks. The thought was that we could prove or disprove the "attraction" hypothesis put forth by Wittemyer et al. (2008) by eliciting a set of core driving concepts / conditions and testing them for agreement using CCA. In so doing, CCA would help to identify whether there is a single set of factors driving migration to all three parks, or whether specific contextual factors specific to each park could be identified in migrants' thinking.
This study followed a three-staged research procedure commonly applied in CCA research (Boster 1986, Romney et al. 1986, Bernard 2017. First, we conducted free-listing to establish the domain. Second, we used pile-sorting to understand participants' categorization of terms within the domain. Third, we gathered ratings data on each of the terms within the domain to determine if there was consensus.

Freelisting
Freelisting is a common, proven, and statistically powerful way of establishing a group's shared knowledge or cognitive domain (Berlin 1992, Brewer 1995, Quinlan 2005. Successive freelisting is also a proven technique for eliciting explanatory models (Ryan et al. 2000). Freelisting with informants provides the "domain" of terms upon which the consensus analysis was subsequently based. Freelisting was conducted during a one-month research trip in May/June 2012. The primary author used convenience sampling and the snowball method to encounter migrants in the main communities in the 10 Km buffer zones of each of the three parks selected. However, during this stage there was neither an attempt to find more specific areas where migrant households were concentrated nor a method for identifying areas more likely to have recent migrants. In total, thirty migrants to each of the buffer zones (N = 90) were interviewed. In that group 55 were men and 35 were women, the average age was 44.4 years old, and they ranged from 21 to 77 years old with a standard deviation of 15.2. Participants had spent average of 11.7 years in their current community. Participants provided a free list of words or phrases These freelists resulted in thousands of terms and concepts, which were re-coded by hand to reduce the number of unique terms and facilitate the analysis of salience and frequency of terms. For example, if multiple terms were used to describe the concept of work (jobs, employment, work, work opportunities), they were condensed to a single concept (work). Appendix 1 shows an example of this transformation for a sample of 30 migrants' responses to the first question. After recoding, the resulting terms for each question were analyzed for salience using Smith's S in Anthropac (4.98). Salience is a measure that combines the frequency of a term in participants' lists and the rank order of the term in the freelist to determine the most relevant terms across the sample (Borgatti 1999, Ribeiro 2012. All terms from each of the questions with a Smith's S result of 0.2 or higher were included in the final domain for the next stage of analysis. In addition, we also used simple frequency of terms to determine whether there was a clear "break" point where responses belonged to a core set of ideas frequently found across participants' freelists (Borgatti 1994). In this case, an "elbow" in data indicating this break appeared at 10% frequency, so all terms found in 10% of participants' freelists were added to the domain. The truncated results of the freelist analysis are available in Appendix 2. As a result of these two steps, we identified an overall domain of 55 terms (see Table 1) deemed to be the most salient in migrants' assessments of community conditions driving movement to the buffer zones of Carara, Arenal, and Barra Honda National Parks.

Pile-sorting and ratings
The next step in the process was to conduct pile-sorting and ratings activities with a new set of participants. Unlike free-listing, our sampling strategy for this stage of data collection refined our focus with the intention of providing a more accurate picture of recent migration trends. The Development Observatory (Observatorio de Desarrollo, OdD) at the University of Costa Rica utilized the 2011 Costa Rican Census to determine the location of recent migrants and provide a sampling strategy for our fieldwork. To do this, the OdD needed to map the park boundaries, map 1 km concentric rings out to 10 km, determine the geostatistical minimum units (GMU, akin to a census tract) included in each ring, and analyze the population growth in each La seguridad (Security) 2 La gente (People) 3 La tranquilidad (Tranquility) 4 El trabajo (Work) 5 El clima (Climate/Weather) 6 La playa (Beach) 7 El ambiente social (Social Environment) 8 El turismo (Tourism) 9 El parque (The Park) 10 Los animales silvestres (Wild Animals) 11 La calidad de vida (Quality of Life) 12 El volcán (Volcano) 13 La paz (Peace) 14 Los ríos (Rivers) 15 Limpio (Cleanliness) 16 Lugar Bonito (Beauty) 17 La inseguridad (Insecurity) 18 Sano (Healthiness) 19 La violencia (Violence) 20 Los servicios (Services) 21 La cultura (Culture) 22 La educación (Education) 23 Las drogas (Drugs) 24 El desempleo (Unemployment) 25 Los precios (Prices) 26 Los servicios médicos (Medical Services) 27 El alcoholismo (Alcoholism) 28 El gobierno (Government) 29 La basura (Garbage) 30 La contaminación (Contamination) 31 El agua (Water) 32 La prostitución (Prositution) 33 Ser céntrico (Centrality) 34 La familia (Family) 35 Las comodidades (Amenities) 36 El transporte (Transportation)  37 El ruido (Noise) 38 Las áreas deportivas (Sports facilities) 39 La agricultura y la ganadería (Agriculture and Livestock) 40 El comercio (Business) 41 La comida (Food) 42 La delincuencia (Delinquency) 43 Sobrepoblación (Overpopulation) 44 Las carreteras (Roads) 45 La pobreza (Poverty) 46 El campo (Countryside) 47 El transito (Traffic) 48 Los robos (Robberies) 49 La economía (Economy) 50 La superación (Improvement / overcoming) 51 La recreación (Recreation) 52 La luz (Electricity) 53 El Banco (Bank) 54 La naturaleza (Nature) 55 El Supermercado (Supermarket) ring based on census data (see Table 2 for measurement of growth per ring). This work then allowed the OdD to provide a sampling strategy for each park that was proportional to both the rings with the most population growth for each park and the overall target of 100 interviews across all three parks (see Table 3). Last, the OdD identified zonas calientes, "hot spots," which were GMUs with higher than average in-migration within the 10 km buffer zone (see Fig. 2 for map).  This "hot spot" approach enabled our small research team to focus fieldwork efforts on areas where it was likely for us to encounter recent migrants. Once we arrived at these hot spots, we often began our search with the proprietors of local stores, pulperias, to help us identify migrant households. Once identified and confirmed, we conducted the interview if they were willing as was indicated by prior, informed oral consent. To find more participants we used a snowball or "chain reference" approach by asking interviewees to identify other potential participants. Pile-sorting and scalar rating interviews were conducted over four weeks in May and June of 2013. The interviews were conducted primarily at places of residence and in Spanish. We interviewed 41 migrants in the buffer zone of Carara, 41 in that of Arenal, and 18 in that of Barra Honda. The demographic characteristics of our sample for each park and all three parks together can be seen in Table 4.
Interviews varied in length, but averaged approximately 30-45 minutes. The first task in the interview was an unconstrained pilesort. All participants were presented with 55 laminated notecards for each of the terms from the domain. To reduce bias from the interviewers or previous participants' sorting, the terms were randomized and assigned a number and each of the participants received the notecards in the same order. In order to be "unconstrained," participants were asked to put the cards into piles / groups in any way that made sense to them, that there was no right answer to the number of piles or how the piles should be organized. Once the participants were satisfied with their sorting, the numbers for the terms in each pile were recorded and the cards were returned to numerical order.
For the second task of the interview, participants were asked to rate the conditions / state of each of the 55 terms (Likert scale of 1-5) first for the community from which they had moved (hereafter referred to as the "previous" community) and then again for their current location within the buffer-zone of the national park (hereafter referred to as the "current" community).
The number 1 represented very poor conditions, 3 average conditions, and 5 indicated very good conditions. Respondents were asked to rate each term for both their previous and current community. For example, we asked "How would you rate the people there (here)?" "¿Como calificaría la gente allí (aquí)?" This wording was somewhat confusing for participants and they were at times unclear if we were asking for how they personally saw the conditions, or how the conditions were perceived in general.
Because we were aware that CCA is dependent on general answers and not individual, personal opinion, we clarified that we sought out ratings "in general" (por lo general).

Pile sorting: MDS and consensus
Pile sort data were compiled and entered into Visual Anthropac and were analyzed using multi-dimensional scaling (MDS), clustering and consensus. The MDS analysis produces a visualization, often referred to as a "cognitive map," of the collective pile-sorting of a sample. Because the research team was interested in attempting to construct a cognitive model of migration for the entirety of migrants to all three parks' buffer zones, MDS was performed on all 100 of the interviewees' pilesorts. The resulting images (see Fig. 3 for MDS by number and Fig. 4 for MDS by term) provide a two-dimensional representation of participants' pile-sorting; the closer together terms are on the MDS the more frequently they were piled together and vice-versa. In addition, hierarchical clustering demonstrates the ways in which participants clustered items in their pile-sorting. As can be seen on the MDS images, five clusters were established: (1) negative / urban (blue); (2) government and business services (pink); (3) essentials for improving oneself (orange); (4) social/community conditions (black); and (5) natural resources and amenities (green).
Because pile-sort data must be converted from its multiple dimensions to fit into a two-dimensional rendering, MDS runs the risk of misrepresenting the data. Anthropac provides a "stress" value to indicate how well the image represents the original data; the closer to zero the stress value the more accurate the map is (Copeland 2011 items to be sorted (Sturrock and Rocha 2000). For our MDS, the stress value was 0.151, which is well below the threshold of 0.372 for a domain of 55 terms as established by Sturrock and Rocha (2000).
Further, Anthropac enables users to test pile-sorting data for consensus by comparing each participants' pile sorts against all others via a factor analysis. As discussed above, the eigenratio should be 3:1 to indicate strong consensus. In our case, the pile sort data produced an eigenratio of 11.647, which indicates a strong fit to the consensus model. This means that there was strong agreement amongst our participants regarding how these terms are related to one another cognitively, and thus it can be assumed that the group shares a cultural understanding about how the terms within the specific domain relate. Put simply, our informants indicated a shared understanding of how to group these terms.

Rating task: consensus analysis
To further test consensus, we used Anthropac (4.98) to analyze the 100 participants' ratings of 53 terms within the domain for both the previous and current community. Upon realizing that the terms "beach" and "volcano" were not found in or relevant to many participants' previous or current location we removed those ratings and conducted our analysis on the remaining 53 terms. We tested for consensus on various subsets of the data in order to understand whether the contexts of individual parks, and/or how the communities from which participants migrated, potentially impacted consensus. First, we tested all of the ratings data for all three parks and for both the previous and current community together. Second, we analyzed the ratings of just the previous and just the current communities for all three parks combined. Third, we conducted analyses on previous, current, and the combined data sets for each of the parks separately.
By parsing the analysis in this way, we intended to differentiate whether these sub-groupings differed in terms of the existence and/or strength of consensus. In so doing, we intended to test the ratings data for the following: (a) consensus around an overall cultural model for the entire sample; (b) consensus around a model for each individual park context; and (c) differences in consensus on ratings of previous versus current communities for both the total sample and for each individual park's buffer zone.
The consensus results are shown in Table 5, which also includes  . 3. Multi-dimensional scaling (MDS) and clusters for all 100 participants' pile sorting. Numbers correspond to numbers assigned to each term found in Table 1. the number of participants, factors (questions), and negative competence scores for each of the analyses. In addition, Appendix 3 provides the culturally correct answer key-what the consensus analysis identifies as the most likely "correct" answer amongst our participants-for the only parsing that meets the standard thresholds for consensus: the ratings of current community conditions by migrants to the buffer zone of Carara National Park.
In brief, consensus analysis performed on the ratings data from all three parks combining both previous and current communities fails to meet the standards for consensus, which means there is not a single, shared cultural model for the entire data set. As well, there is no consensus in our data when results for the entire sample from all three parks are parsed and analyzed by previous or current community (e.g., "All Parks Previous" or "All Parks Current" in Table 5). When the data is broken down by individual park, there is also no consensus when both previous and current community ratings are combined (e.g., "Carara combined"). In addition, all of these analyses have at least one negative competence score, and some of them have many negative scores, which further demonstrates the lack of a single cultural model. When the individuals with negative scores were identified, there were no obvious or discernable patterns amongst them (e.g., age, occupation, gender). Although most of the negative scores were found in the 1 km rings furthest from the parks (km 9 and km 10), each parsing contained several individuals at the same distance that had positive competence scores. Put simply, our results indicate that there is little agreement and not a single cultural model of migration to national park buffer zones amongst our sample of migrants in Costa Rica.
However, our analysis does indicate that there is evidence for consensus amongst our informants in their ratings of conditions in their current community within the buffer zone of Carara. In addition, the current community ratings for Arenal National Park are very close to meeting standard thresholds discussed above.
With an average competence score of 0.49, Arenal informants' ratings come very close to the threshold of 0.5, but they do not quite meet 3:1 eigenvalue ratio expected. It is tempting to say that the Barra Honda results are approaching the standards for consensus. However, they do not because of the more stringent standards required for a sample size below 30 individuals (Weller 2007). In sum, the ratings data indicate that, for one out of three parks, there was a shared cultural model among recent migrants regarding their current community.

Ratings tasks: statistical comparison of previous and current
The lack of consensus was somewhat surprising considering the consistency in explanations the research group knew qualitatively from discussions with migrants. Thus, we subjected the ratings data to a separate statistical analysis to see whether there were identifiable and significant trends in migrants' assessments of these 53 terms for previous and current communities. In order to determine which ratings demonstrated statistically significant differences between prior and current communities, we compared the previous and current ratings data of the 53 terms for the entire sample. In this analysis, we used the non-parametric Wilcoxon signed-rank test (Wilcoxon 1945). Because this approach involves multiple comparisons, we applied a Bonferroni correction (Bonferroni 1936) to the significance level, which reduces the probability of a Type I error.
We found that 23 of the variables had a statistically significant change in rankings (marked with a * in Table 6). Overall, this analysis shows that many terms that are associated with the negative realities of urban living were not only improved, but in a statistically significant way. Thus, migrants indicated significant improvement in conditions like insecurity, contamination, delinquency, traffic, contamination, noise, violence, and overpopulation. The same can be said for most terms associated with the ability to live life with peace, tranquility, beauty, and security. Third, terms associated with rural life and livelihoods (nature, beauty, animals, rivers, agriculture and livestock, tourism, and the park) are also significantly improved.
Interestingly, a number of terms that were generally rated as improved when looking simply at their means were not different in a statistically significant way. Many of these are related to social ills that are not unique to urban settings such as poverty, drugs, prostitution, and alcoholism. As well, the lack of a significant https://www.ecologyandsociety.org/vol27/iss4/art16/ Fig. 4. Multi-dimensional scaling (MDS) and clusters for all 100 participants' pile sorting labeled with the actual terms from Table 1. difference between ratings on terms like garbage and roads indicates that the perception is that these government services are poorly managed across the country. Last, the lack of a statistical difference in their ratings of healthiness, cleanliness, social environment and people indicates that migrants see these socioenvironmental conditions as stable across the country, that buffer zones of national parks do not provide a difference in these areas. However, the fact that there is a statistically significant difference in ratings on the overall quality of life indicates that park buffer zones do provide an overall sense of improved life conditions that are attractive to Costa Rican migrants.
Our results show that only four terms were perceived as significantly worse in buffer zone communities (see Table 6). These negative ratings for three of these terms are found in services that are impacted by geographic isolation and/or the integration of these communities with tourism-based economies. First, prices were seen as significantly worse, which is a phenomenon that is frequently associated with tourism-based economies and isolation. Further, migrants perceived that access to transport was significantly worse in the buffer zones. Third, migrants rated medical services as significantly poorer, which is again the reality for rural Costa Rica especially in comparison to the facilities and treatments available in urban and suburban San José. Last, climate was the one term that was rated as significantly poorer. This is attributable to the fact that all three parks were in warm, lowland climates that were being compared to the urban, highland, and cool areas like the capital of San José from which many migrants had moved.

DISCUSSION
The findings uncovered via our work in Costa Rica contribute important, migrant-centered, contextual data that illuminates what factors drive migration decisions to PA buffer zones. Our research was designed to use a CCA approach to see whether there is consensus amongst migrants' motivations, which could be further used to test the singular "pull" model put forth by Wittemyer et al. (2008). In so doing, our work adds the following to the conversation regarding PAs and human migration: (1) CCA results further complicate the hypothesized direct links between PAs, integrated conservation and development, and population growth proposed by Wittemyer et al. (2008); (2)  As was stated at the outset of this paper, some studies have criticized the methods by which Wittemyer et al. (2008) derived their evidence, as well as their hypothesized, generalizable, worldwide drivers for the results they observed (Joppa et al. 2009, Joppa 2012. Because of the lack of consensus in our study, we join others in demonstrating that, even within a single country, the social, political, economic, and ecological contexts surrounding every park are unique and must be taken into account when explaining the patterns of human population growth found there (Fay 2011, Hoffman 2011, Guerbois et al. 2013, Hartter et al. 2015. Importantly, our deployment of cultural consensus analysis offered a unique perspective from which to assess migrant motivations. Disagreement shown in the lack of cultural consensus for previous and current community conditions across the entire sample suggests that a singular explanation or causal mechanism for migration to PA buffer zones is unlikely. In fact, the lack of consensus within almost all our analytical subsets, apart from the Carara "current" community, can be seen as evidence that singular explanations are unhelpful. The lack of consensus for the combination of both previous and current communities is likely due, in part, to the variation in previous communities from which migrants arrived to the parks' buffer zones. Although there were a few locations that "sent" multiple migrants, both our overall sample for each park and for individual parks showed incredible diversity in sending communities. Clearly, the diversity of previous communities impacted the consensus results for the "previous"-only analyses. This same variation in ratings of their previous communities is what likely lead to the lack of consensus for the "combined" analyses. Ultimately, the near complete lack of ratings consensus shows that despite some level of agreement among migrants (seen in the free listing, pile sorting, and comparison of means results), migration to PA buffer zones is complex and contradictory.
The statistical analysis of respondents' ratings illuminates those terms found within the domain that are seen by migrants as notably different between their previous and current communities. First, a number of the negative elements usually associated with urban living like overpopulation, noise, violence, delinquency, poverty, and insecurity are rated as improved. Second, themes associated with the ability to live life with peace, tranquility, beauty, cleanliness, and security are all seen to be improved in their current communities. Not surprisingly, many of these themes are the opposite of negative conditions that our participants frequently associated with their previous, often urban, communities. Third, terms associated with the rural life and livelihoods based on natural resources (nature, beauty, animals, rivers, agriculture and ranching, tourism) were rated as significantly better. Statistical analysis shows that those terms with positive improvement can be interpreted as "pull" conditions in the current buffer zones, especially when seen in relation to their previous communities and the "push" conditions found in those locations. Overall, this provides insight into the conditions that migrants have sought out and support the earlier conclusion particular to Costa Rica (Hoffman 2011, Hoffman 2020) that, despite some sacrifices in convenience and services, the peace, security, natural amenities, and overall quality of life in the buffer zones play an important part in the attraction of migrants to their current communities.
Overall, the statistical analysis of the overall ratings between previous and current conditions shows migrants saw some elements typically associated with conservation and development projects (i.e., tourism) as a draw to their current community. However, a large number of the significant terms have little to do with the opportunities directly provided by PA-based development and, therefore, contradicts the explanation put forth by Wittemyer et al. (2008). Thus, our work further supports the conclusion of Guerbois et al. (2013), that it is critical to approach this question with methods and analyses that document people's livelihoods, histories, education, perceptions of conservation and PAs, and the natural resources used in each context. Thus, we reinforce the need to combine census and demographic data with social science methods such as surveys, interviews, and focusgroups to disentangle the localized complexity of migration (Salerno et al. 2014, Hartter et al. 2015. We add to this by not only combining geospatial and demographic analyses with traditional social science research methods but also providing insight into the viability of employing consensus analysis as a tool for understanding and uncovering migrants' motivations. The consensus data derived from our targeted sampling strategy for pile sorting and rating tasks supports the suggestion that selecting an appropriate scale of analysis is critically important because of the high levels of variability within all of these potential drivers that exist in the buffer zones of individual parks (Salerno et al. 2014, Salerno 2016). The collaboration with OdD and targeted "hot spot" methodology allowed us to focus our interviews in the places where migration was most prevalent rather than being distracted by existing population centers or anecdotal evidence. By answering Salerno et al.'s call for employing appropriate scales as determined by a combination of geospatial and demographic analyses as the basis for our work, we were able https://www.ecologyandsociety.org/vol27/iss4/art16/ to disentangle the context-dependent relationship between PAs and migration. In so doing, our work further supports Gupta's (2015) point that conclusions about migration and PAs are dependent on the scale at which data is analyzed. Our analysis concurs with her suggestion that even when a 10 km-scale shows growth, localized analyses show more variability. Indeed, the lack of consensus on current community conditions points out that there are important differences in migrants' evaluations even within a single park's buffer zone communities.
Furthermore, the methodology employed here provides rare insight into the specific mindsets of recent migrants that further illuminates the contextual complexities of the PA-migration nexus. Our study is relatively unique in its explicit focus on recent migrants to the PA buffer zones (cf. Salerno 2016). Many studies survey or interview all groups living on the edges of PAs (e.g., Guerbois et al. 2013, Bamford et al. 2014) and/or use interview data from elders to elicit thoughts on how buffer zones have changed over longer periods of time (e.g., Hartter et al. 2015). Others, such as Levang et al. (2012) used targeted methods to find areas that are likely to have new migrants, but deployed interviews and questionnaires among both recent and longer term, secondgeneration immigrants to PA buffer zone. Although there is no doubt that these studies were able to elicit important information regarding the conditions within PA buffer zones and their relationship to population growth and in-migration, they do not concentrate specifically on relatively recent migrants and their perceptions of buffer zone social, economic, political, and environmental conditions.
Our work joins Salerno (2016) in engaging migrants via mixed field methodologies with the explicit intent of eliciting a model to understand, as well as potentially predict, what motivates migrant decisions. The fact that our data point to factors such as tranquility, safety, and factors outside of resources typically associated with the natural or development-induced resources corroborate Salerno's findings that migrant decision making to PA buffer zones is driven by a complex mix of factors that go beyond simplistic anecdotes about the attraction of land and natural resources. Our work parallels Salerno (2016) in the use of free-listing with actual migrants to elicit the primary categories driving migrant decisions. Salerno's (2016) findings offer a very different picture of migration drivers for the radically different context of western Tanzania, which further demonstrates the need for the types of studies we have conducted.
Last, the demographic and mapping work carried out by the OdD supports the critiques of the demographic methods and data sets used by Wittemyer et al. (Joppa et al. 2009, Joppa 2012). We did not set out with the specific intent of analyzing the spatial relationship between population growth and PA borders, but the OdD analysis further supports Joppa et al.'s (2009) critical re-analysis of the demographic and spatial data in regard to the relationships between parks and migration-driven population growth. As Joppa et al. (2009) point out, if the park and its resources / development were the main draws for population growth one would expect that the areas of greatest population growth would be in closer proximity to the park boundaries. Joppa et al.'s (2009) re-analysis observed that growth could more likely be an outgrowth of the population expansion of nearby towns and urban centers. Our finer scale census analysis based on GMU supports this position by showing that inmigration numbers and "hot spots" are not located within the first few kilometers of the park boundaries (see Table 2 and Fig.  2) for the chosen Costa Rican parks. Instead, the areas of greatest migrant growth were located between 3-5 km and 9-10 km from the park boundaries. This is in part driven by the conditions of the buffers closest to the parks in that they tended to be the least https://www.ecologyandsociety.org/vol27/iss4/art16/ developed, had fewer public services, and were likely to be highly sloped lands unattractive for agriculture. Ultimately, this further complicates the notion that it is access to the parks' resources, natural or infrastructural, that drives migration.

CONCLUSION
Overall, our research adds further evidence to existing discussions of how to measure the effects of PAs on migration and what evidence would be expected if PAs and their resources were the primary driver of migration decisions. The analyses presented contribute to larger discussions about PAs, human migration, and the effectiveness of conservation and development policies as a strategy for biodiversity conservation. Costa Rican migrants' perceptions of conditions in the buffer zones of three Costa Rican National Parks both supports and contradicts the connections between PAs, conservation and development, and human migration put forth by Wittemyer et al. (2008). Certain conditions in buffer zones were consistently and significantly rated higher by migrants and could be interpreted as conditions that attracted migrants. However, it is critical to note that it seems as if migration decisions were often related to conditions and resources that are not directly produced or provided by the PA or conservation and development policies. Thus, our analyses supports existing qualitative work in these contexts that question whether conservation and development in Costa Rica creates population growth in buffer zones due to in-migration (Hoffman 2011, Dehler 2015, Arends 2017, Hoffman 2020. In so doing, this work provides important contextualization of the relationship between PAs, migration, and population growth, as was suggested by Wittemyer et al. (2008). More specifically our work provides empirical data in response to the call for "more real-world examples of immigration to protected areas ... along with information on the reasons for the immigration and the benefits that may be provided by the protected area, such as income or natural resources" (Bamford et al. 2014:504).
This study reinforces the conclusions of many scholars regarding the need to base conservation policies upon the specific and varying social-ecological conditions found within the buffer zones of individual PAs. We concur with the others that there is a need to base conservation decisions on the political ecology, policy, and socioeconomic factors and history of individual PA development (Bamford et al. 2014, Levang et al. 2012, Salerno et al. 2014, Hartter at el. 2015. Unfortunately, many PA administrators, national conservation authorities, and NGOs often do not have the skills or the resources, especially time and labor, to conduct such fine-grained analyses in order to understand the ways in which PAs (and other local factors) affect migration and have potential impacts upon the biodiversity they are established to protect. Instead, they are forced to rely on generalizations and anecdotal information to guide policy decisions.
In sum, our work further stresses the need to engage with and understand the impacts of conservation efforts upon local people as has been repeatedly emphasized by social scientists (West et al. 2006, Büscher and Fletcher 2019, Agrawal et al. 2021. This is particularly relevant in light of the forceful debates about creating more extensive conservation efforts to combat the continued decline of global biodiversity such as the Half Earth (Wilson 2016) or protecting 30% for nature proposals (Waldron et al. 2020). Our methods provide another potential pathway (cf. Salerno 2016) for analyzing how PAs interact with local social, economic, and political conditions to impact migrant decision making that are critical for understanding the trade-offs necessary for attending to the needs of both biodiversity and local human populations.
Responses to this article can be read online at: https://www.ecologyandsociety.org/issues/responses. php/13529