Understanding fire conflict through stakeholder mapping in Madagascar's grassy Biomes

This dataset contains fuzzy cognitive mapping (FCM) data examining stakeholder perspectives on landscape fire drivers in Madagascar's Ambatofinandrahana district. The data were collected through 28 focus groups (n = 133 participants) conducted in April-May 2023 across five stakeholder groups: government officials, conservation practitioners, community leaders, rural farmers, and rural herders. Participants collectively identified variables influencing landscape fires and mapped causal relationships between them, assigning directional connections and strength ratings (1-4 scale). The repository includes: (1) an aggregated stakeholder matrix consolidating all 28 focus group maps; (2) five stakeholder-level aggregated matrices used for hierarchical clustering analysis; (3) a consolidated variable list documenting how 200+ original variables were condensed into 36 analytical categories; (4) network metric summary tables for variables including Katz centrality scores, in-degree, out-degree, OD/ID ratios; (5) variable-to-variable relationship data for each stakeholder group. These data support the findings reported in Convery-Fisher et al., "Understanding Fire Conflict Through Stakeholder Mapping in Madagascar's Grassy Biomes," published in People and Nature. # Understanding fire conflict through stakeholder mapping in Madagascar's grassy Biomes Dataset DOI: [10.5061/dryad.5mkkwh7jk](doi.org) ## Data Description ### Research Context and Objectives These data were collected as part of a participatory research effort to understand and compare stakeholder perspectives on landscape fire drivers in Madagascar's grassy biomes. Fire management in Madagascar represents a long-standing conflict between rural communities who use fire as a traditional land management tool and government and conservation authorities who view fire as environmentally destructive. Despite decades of fire suppression policies, tensions persist, and management approaches remain ineffective. This research employed fuzzy cognitive mapping (FCM)—a semi-quantitative participatory method—to systematically capture how different stakeholder groups perceive the causal relationships driving landscape fires in the Ambatofinandrahana district, central Madagascar. The study was designed to: 1. Identify which variables stakeholders perceive as most important in influencing landscape fire dynamics 2. Compare how stakeholder perceptions of causal relationships vary across groups 3. Reveal areas of agreement and divergence that could inform collaborative fire management approaches ### Research Design Between April and May 2023, we conducted 28 focus group discussions with 133 participants representing five key stakeholder groups: government officials (n = 15), conservation practitioners (n = 18), community leaders (n = 52), rural farmers (n = 25), and rural herders (n = 23). Focus groups were held across nine rural villages and two towns in the Ambatofinandrahana district. During structured 2-hour sessions, participants collectively identified social, economic, environmental, and political variables they perceived as influencing landscape fires. They then mapped causal relationships between these variables, creating visual cognitive maps that represent their shared understanding of fire system dynamics. Each relationship was assigned a direction (positive or negative) and strength rating (1-4 scale), providing semi-quantitative data suitable for network analysis. The resulting dataset captures diverse and sometimes conflicting mental models of fire dynamics, revealing not just differences in priorities but fundamental divergences in how stakeholders understand cause-and-effect relationships in fire management. These data provide empirical grounding for understanding why fire policies often fail and where opportunities for collaboration may exist. ### Study Significance This dataset represents one of the first systematic, comparative analyses of stakeholder perspectives on fire management in Madagascar using structured participatory methods. The fuzzy cognitive mapping approach bridges qualitative and quantitative traditions, producing data that preserve the contextual richness of local knowledge whilst enabling statistical comparison across stakeholder groups. The methods and analytical approaches demonstrated here are transferable to other natural resource conflicts globally where diverse perspectives complicate governance. --- **Related Publication:**\ Convery-Fisher, E.D., Devenish, A., Staddon, S., Rafanomezantsoa, F.L., & Lehmann, C.E.R. Understanding Fire Conflict Through Stakeholder Mapping in Madagascar's Grassy Biomes. *People and Nature* [full citation upon publication]. **Funding:**\ This research was supported by the University of Edinburgh, Royal Botanic Garden Edinburgh, NERC E4 Doctoral Training Programme (Grant No. NE/S007407/1), Albert Reckitt Award (Royal Geographical Society with IBG), and UK International Development via the Biodiverse Landscapes Fund project 'Achieving Sustainable Forest Management through Community Protected Areas in Madagascar' (ecm 62237). **Geographic Coverage:**\ Ambatofinandrahana district, Amoron'i Mania Region, central Madagascar **Temporal Coverage:**\ April-May 2023 ### Files and variables ## File: aggregated_matrix_final.csv **Description:** Adjacency matrix representing the aggregated cognitive map across all 28 focus groups and five stakeholder categories. Each cell contains the weighted connection strength between variables (row to column). Positive values indicate positive causal relationships (increase in row variable causes increase in column variable), negative values indicate negative relationships (increase in row variable causes decrease in column variable). Values range from -4 to +4, with 0 indicating no perceived connection. This matrix was used to generate network metrics and visualizations in Figures 3 and 5. **Matrix structure:** Variables as both rows (source) and columns (target), with cell values representing connection weights. **Missing values:** Represented as 0 (no connection perceived by any stakeholder group) ### Variables (n = 36): * **accidental_fire:** Unintentional fire escapes from cooking, land preparation, or waste burning * **advantage_fire:** The functional benefits of using fire as a land management tool (consolidated from multiple perceived benefits) * **agricultural_fire:** Use of fire to clear land for cultivation or prepare planting areas * **agricultural_inputs:** Availability of farming resources such as fertilisers, pesticides, tools, and seeds * **arson:** Intentional destructive fire use, typically for conflict, retaliation, or criminal purposes * **lack_of_basic_services:** Inadequate provision of education, healthcare, markets, and government services in rural areas * **cattle_ranching:** Livestock management practices and the economic importance of cattle herding * **cheap:** Fire as a low-cost or no-cost land management alternative (financial accessibility of fire use) * **climate_phenomena:** Seasonal weather patterns, rainfall timing, drought conditions, and temperature affecting fire behaviour * **community_conflict:** Social tensions within or between communities related to resource use and fire management * **community_organisations:** Local institutions such as *Vondron'Olona Ifotony* (VOI), fire management committees, and community-based natural resource management groups * **cultural_fire:** Traditional fire use practices embedded in cultural identity and customary knowledge systems * **dahalo_presence:** Cattle rustling activity; theft of livestock often linked with fire as a diversion tactic (from Malagasy term *dahalo* = cattle rustler) * **demographics:** Population size, density, growth, age structure, and migration patterns * **different_perspectives:** Divergent viewpoints and knowledge systems among stakeholders regarding fire use and management * **easily_available:** Accessibility and availability of fire as a tool (ease of ignition) * **enforce_rights:** Legal recognition and enforcement of land tenure, customary rights, and resource access rights * **expansion_of_agriculture:** Increasing agricultural land area through conversion of grasslands or forests * **fire_properties:** Fire behaviour characteristics including spread rate, intensity, timing, and extent * **forest_conservation:** Protection efforts targeting forest ecosystems, including protected area management * **forest_products:** Non-timber forest resources including firewood, charcoal, building materials, and medicinal plants * **grassland_fire:** Landscape fires burning in grassland ecosystems (outcome variable of interest; also called *dorotanety* in Malagasy) * **grazing_fire:** Deliberate use of fire to regenerate or manage grazing land; renew pastures and improve forage quality for livestock (also called pasture burning in manuscript) * **habit:** Habitual or customary fire use; the use of fire more out of tradition or routine than reasoned decision-making * **lack_of_concern:** Insufficient attention or care regarding fire impacts, fire risk, or fire management responsibilities * **law_enforcement:** Government efforts to implement and enforce fire regulations, including fines, arrests, and prosecutions * **land_degredation:** Environmental degradation including soil erosion, loss of vegetation cover, and declining ecosystem health (also called environmental degradation in manuscript) * **lightening:** Natural ignition source; lightning strikes causing fire ignition * **lighters:** Availability of ignition tools (matches, lighters, etc.) * **local_fire_management:** Village-level and community-led efforts to control or extinguish landscape fires (also called community fire suppression in manuscript) * **local_infrastructure:** Quality of roads, bridges, transport networks, communication systems, and market access in rural areas (also called inadequate rural infrastructure in manuscript) * **national_political_context:** Broader political dynamics, policy changes, and governance structures affecting fire management * **ngo_support:** Presence and activities of non-governmental organisations providing conservation, development, or capacity-building support * **people_passing:** Transient individuals traveling through the landscape who may inadvertently or deliberately start fires * **pests:** Agricultural and livestock pests targeted for control through burning * **pyrophilic_species_presence:** Presence of fire-adapted or fire-promoting plant species in the landscape * **qualities_of_the_natural_environment:** Landscape conditions including vegetation type, fuel load, topography, and ecosystem characteristics (also called landscape conditions in manuscript) * **social_cohesion:** Strength of social bonds, trust, and cooperation within communities * **social_status:** Prestige and social standing associated with cattle ownership and herd size (also called prestige of cattle ownership in manuscript) * **state_presence:** Government capacity including staffing, resources, and authority to implement fire management (also called limited government capacity in manuscript) * **time:** Seasonal timing of fire use and fire seasons (also called seasonal weather patterns in some contexts) **Note:** Some variables appear under alternative names in the published manuscript to improve clarity for readers; the mapping between dataset variable names and manuscript terminology is indicated in parentheses above. **Description:** Documentation showing how 200+ original variables identified by participants during focus groups were consolidated into 36 analytical categories. This two-sheet workbook provides complete transparency in the variable consolidation process, showing both the consolidation decisions and the final variable definitions used in analysis. ## File: condensing_variables.xlsx **Description:** Documentation showing how 200+ original variables identified by participants during focus groups were consolidated into 36 analytical categories. This two-sheet workbook provides complete transparency in the variable consolidation process, showing both the consolidation decisions and the final variable definitions used in analysis. ### Sheet 1: Consolidation_Mapping This sheet documents the consolidation process, showing how each original variable identified by participants was grouped into broader analytical categories. **Variables:** * **original_variable:** Variable name or phrase as stated by participants during focus groups (in English, French, or translated from Malagasy). Examples include specific terms like "cow_farming", "firebreak", "road_condition", etc. * **consolidated_variable:** The final analytical category name assigned to this original variable (matches variable names in aggregated_matrix_final.csv and other datasets). Total of 36 consolidated categories. * **rationale_notes:** Explanation for the consolidation decision, describing why this original variable was grouped into its assigned category (e.g., "Climate/weather variable", "Government institutional capacity indicator", "Social issue reflecting inadequate community support systems") **Number of rows:** 200+ original variables consolidated into 36 categories **Purpose:** This sheet demonstrates the semantic and conceptual logic used to reduce the complexity of participant-generated variables whilst preserving the meaning and diversity of perspectives. Two researchers (ECF and LF) independently reviewed all consolidation decisions to ensure accuracy and minimize information loss. ### Sheet 2: Final_Variable_Definitions This sheet provides the complete list of consolidated variables with their manuscript names and detailed definitions used throughout the analysis and publication. **Variables:** * **consolidated_variable:** The analytical category name used in the dataset files (e.g., "accidental_fire", "cattle_ranching", "dahalo_presence"). These names match those used in all matrix files and network analysis outputs. * **New Variable Name:** The terminology used in the published manuscript for clarity and readability (e.g., "accidental fires", "cattle ranching", "cattle rustling"). Some variables retain the same name; others were adapted for a general scientific audience. * **Explanation:** Comprehensive definition of what this variable represents in the context of fire management in Madagascar. Definitions include the scope of the concept, relevant examples, and contextual information to aid interpretation. **Number of rows:** 36 consolidated variables **Purpose:** This sheet serves as the definitive reference for understanding what each variable means, bridging between the dataset terminology and the published manuscript terminology. Users of the dataset should refer to these definitions when interpreting results. --- **Note on terminology:** Some variables appear under different names in the dataset files versus the manuscript to balance database naming conventions (e.g., underscores, brevity) with reader accessibility (e.g., descriptive phrases). Sheet 2 provides the authoritative mapping between these naming systems. ## File: network_metrics_summaries.xlsx **Description:** Summary tables of key network metrics for variables in the fire management cognitive mapping system. This three-sheet workbook presents the most influential variables identified through network analysis of the aggregated stakeholder matrix. These summaries support the identification of system drivers (transmitter variables), system outcomes (receiver variables), and overall variable importance reported in the Results section and Tables 2-3 of the manuscript. **Missing values:** Not applicable (only variables meeting specific thresholds are included in each sheet) ### Sheet 1: Katz_Centrality This sheet presents the top 10 most influential and connected variables in the aggregated network, ranked by Katz centrality scores. These variables represent the core components of the fire management system as perceived across all stakeholder groups. **Variables:** * **Absolute Rank:** Numerical ranking from 1 (highest centrality) to 10 (1-10) * **Variable:** Variable name as appears in the published manuscript (may differ slightly from dataset terminology; see condensing_variables.xlsx Sheet 2 for mapping) * **Type:** [Column appears in header but values not shown - likely variable classification as Transmitter/Receiver/Ordinary] * **Katz Centrality:** Katz centrality score calculated with attenuation factor α = 0.1. Higher scores indicate greater overall importance and connectivity within the network. Scores are unitless; range shown is 1.04-2.53. **Number of rows:** 10 variables (top-ranked by centrality) **Purpose:** Identifies the most structurally important variables in the fire system. "Uncontrolled Grassland Fires" (the outcome of interest) has the highest centrality, followed by direct fire practices (pasture burning, arson, agricultural fire) and key contextual factors (infrastructure, law enforcement, cattle rustling). **Note:** The manuscript refers to "landscape fires" or "grassland fires" rather than "Uncontrolled Grassland Fires," but these terms are synonymous in this context. --- ### Sheet 2: OD_ID_RATIO This sheet identifies transmitter variables—those that primarily influence other components of the system rather than being influenced themselves. Transmitter variables can be thought of as upstream drivers or root causes in the causal network. **Variables:** * **Rank:** Numerical ranking based on classification category and out-degree to in-degree ratio * **Variable:** Variable name as appears in the manuscript * **Type:** Classification of transmitter type: * **'True' transmitters:** Variables with high out-degree and zero or near-zero in-degree (pure drivers with no incoming connections). Numbers in this column represent out-degree centrality scores. * **'Ordinary' transmitters:** Variables with non-zero in-degree but still functioning primarily as drivers (out-degree substantially exceeds in-degree). Listed with their rank within this category in parentheses. * **OD Centrality / OD/ID Ratio:** * For 'True' transmitters: Out-degree centrality score (cumulative strength of outgoing connections) * For 'Ordinary' transmitters: Out-degree to in-degree ratio (how many times greater the out-degree is compared to in-degree) **Number of rows:** 15 transmitter variables total (5 'True', 10 'Ordinary') **Classification criteria:** Variables were classified as transmitters if they had zero in-degree with non-zero out-degree, or if their OD/ID ratio fell in the upper quartile of all variables. **Purpose:** Identifies root causes and external forcing factors in the fire system. 'True' transmitters represent fundamental drivers that are not influenced by other variables in the system. 'Ordinary' transmitters are influenced by other factors but still function primarily as causes rather than effects. **Key findings:** Socio-economic factors (community organisations, agricultural inputs, poor local services), institutional factors (political instability, NGO support), and biophysical conditions (fire properties, environmental conditions, climate) emerge as primary drivers. --- ### Sheet 3: ID_OD_RATIO This sheet identifies receiver variables—those that are primarily influenced by other components of the system rather than influencing others. Receiver variables can be thought of as outcomes or consequences in the causal network. **Variables:** * **Rank:** Numerical ranking based on classification category and in-degree to out-degree ratio * **Variable:** Variable name as appears in the manuscript * **Type:** Classification of receiver type: * **'True receivers':** Variables with high in-degree and zero or near-zero out-degree (pure outcomes). Listed with in-degree centrality score. * **[Ordinary receivers]:** Variables with non-zero out-degree but still functioning primarily as outcomes (in-degree exceeds out-degree). Rank within this category shown in parentheses. * **ID Centrality / ID/OD Ratio:** * For 'True receivers': In-degree centrality score (cumulative strength of incoming connections) * For other receivers: In-degree to out-degree ratio (how many times greater the in-degree is compared to out-degree) **Number of rows:** 9 receiver variables (1 'True' receiver, 8 ordinary receivers) **Classification criteria:** Variables were classified as receivers if they had zero out-degree with non-zero in-degree, or if their ID/OD ratio fell in the upper quartile of all variables. **Purpose:** Identifies system outcomes and consequences. These variables represent the effects or results of other system components rather than causes. **Key findings:** "Uncontrolled Grassland Fires" is the primary 'True receiver' with the highest in-degree centrality (16.4), confirming it functions as the main system outcome. Other receivers include specific fire types (arson, pasture burning, agricultural fires, accidents) and management responses (law enforcement, local fire management), showing these are consequences of underlying drivers. --- **Interpretation notes:** 1. **Transmitter vs. Receiver classification:** A variable can appear in multiple sheets if it has moderate values for both in-degree and out-degree. For example, "Community Organisations" appears as both a transmitter (OD/ID ratio context) and receiver (ID/OD ratio context), indicating it both influences and is influenced by other system components. 2. **Terminology alignment:** Variable names in these sheets use manuscript terminology for readability. To find the corresponding dataset variable name (as used in matrix files), refer to condensing_variables.xlsx Sheet 2. 3. **Quartile thresholds:** The specific OD/ID and ID/OD ratio values that define the "upper quartile" cutoffs for transmitter/receiver classification are available in the full network analysis scripts. 4. **Centrality vs. Frequency:** High centrality indicates structural importance in the causal network (how strongly a variable connects to others), not how frequently participants mentioned the variable. ## File: stakeholder_variance_analysis.csv **Description:** Top 10 most important variables for each of the four stakeholder clusters identified through hierarchical clustering analysis. This file shows cluster-specific Katz centrality rankings used to generate Table 3 and Figure 5 in the manuscript, revealing how different clusters prioritize different fire drivers. **Missing values:** Not applicable (only top 10 variables per cluster are included) **File structure:** Long format with one row per variable per cluster (40 total rows: 4 clusters × 10 variables each) ### Variables: * **variable:** Variable name as it appears in the manuscript (e.g., "Habitual fire use", "Pasture burning", "Environmental degradation"). To find corresponding dataset variable names, refer to condensing_variables.xlsx Sheet 2. * **katz_centrality:** Katz centrality score for this variable within this specific cluster's aggregated cognitive map. This is not the overall centrality across all stakeholders, but rather the importance of this variable for this particular cluster. Scores range from approximately 1.03 to 1.55. Higher scores indicate greater perceived importance and connectivity within that cluster's understanding of fire dynamics. * **cluster_number:** Cluster assignment (1-4) from hierarchical clustering analysis (Figure 4): * **Cluster 1** (n = 9 focus groups): Government officials (40%) and conservation practitioners (40%), with some community leaders (20%) * **Cluster 2** (n = 4 focus groups): Community leaders (75%) and conservation practitioners (25%) * **Cluster 3** (n = 5 focus groups): Exclusively community leaders (100%) * **Cluster 4** (n = 10 focus groups): Rural farmers (50%) and herders (50%) **Number of rows:** 40 (10 variables per cluster × 4 clusters) **Purpose:** This file demonstrates how stakeholder clusters differ in their perceptions of fire drivers. By comparing the top 10 variables across clusters, we can identify areas of agreement (variables that appear in multiple clusters' top 10) and divergence (variables highly ranked by one cluster but absent from others). For example: * "Habitual fire use" ranks #1 for Clusters 1 and 2 but doesn't appear in Cluster 4's top 10 * "Pasture burning" appears in all four clusters but with varying centrality scores * "Cattle rustling" appears only in Cluster 4's top 10 **Relationship to figures/tables:** * Data used to create **Table 3** showing top variables per cluster * Used to generate **Figure 5** comparing variable importance across clusters * Cluster assignments reference **Figure 4** (hierarchical clustering dendrogram) **Note on variable terminology:** Variable names in this file use the manuscript terminology for readability (with spaces and capital letters). To find the corresponding variable name used in other dataset files (e.g., aggregated_matrix_final.csv), refer to the mapping in condensing_variables.xlsx Sheet 2. # Other Publicly Accessible Locations and Data Sources ## Other publicly accessible locations of the data: * **None.** This Dryad repository is the primary and sole public repository for these data. Upon publication of the associated manuscript in *People and Nature*, a link to this Dryad dataset will be included in the Data Availability Statement. ## Data was derived from the following sources: * **None.** These data represent original primary data collected through field research conducted in Madagascar in April-May 2023. The data were not derived from, extracted from, or based upon any existing datasets or secondary sources. **Methodological Framework:** Whilst the data collection methodology follows established fuzzy cognitive mapping protocols as described in the literature (particularly Özesmi & Özesmi 2004; Devisscher et al. 2016; Tebbutt et al. 2021), all data values, variables, and relationships were generated directly by research participants during focus group discussions and represent their original perspectives and knowledge. ## Human subjects data All participants provided written or oral prior informed consent before participating in focus group discussions, in compliance with ethical approval from the University of Edinburgh School of GeoSciences' Research Ethics and Integrity Committee (approval #GEOS2022-585). Participants were informed that anonymised and aggregated data from the focus groups would be made publicly available to support scientific transparency and enable replication of research findings. Consent was recorded for all 133 participants across 28 focus groups conducted in April-May 2023.