Data from: Whose success counts? Reconciling stakeholder goals in Madagascar's reforestation policy
Data files
Nov 10, 2026 version files 128.14 KB
-
Centrality_Government.csv
7.42 KB
-
Centrality_Leader_Community.csv
5.81 KB
-
Centrality_Local_Community.csv
12.13 KB
-
Centrality_Metrics.csv
17.06 KB
-
Centrality_NGO.csv
4.06 KB
-
harmonised_factor.csv
13.57 KB
-
Louvain_Communities.csv
4.06 KB
-
Matrix_Consistency_Check.csv
5.22 KB
-
R_Code.R
27.21 KB
-
README.md
30.53 KB
-
Top_10_Degree_Centrality.csv
1.07 KB
Abstract
This dataset supports the article “Whose success counts? Reconciling stakeholder goals in Madagascar’s reforestation policy”. It contains de-identified and aggregated data derived from 62 semi-structured interviews, 101 focus-group discussions and 101 fuzzy cognitive mapping exercises conducted between 2024 and 2025 across four protected-area landscapes in Madagascar: Ambohitantely, Ankafobe, Ankarafantsika and Andasibe. The research involved 333 participants representing government and local authorities, non-governmental and technical organisations, community leaders and local community members.
The dataset documents stakeholder perceptions of the ecological, social, technical, economic and institutional factors influencing reforestation outcomes. It includes a summary of participant numbers by landscape and stakeholder group; a lookup table linking 442 participant-defined raw factors to 191 harmonised factors; signed and weighted fuzzy cognitive mapping data in which causal relationships range from −4, indicating a very strong negative relationship, to +4, indicating a very strong positive relationship; stakeholder-level aggregated adjacency matrices; and Katz centrality scores and rankings used to identify structurally prominent factors within each stakeholder network. Where no relationship between two factors was identified, the corresponding value is zero. Analysis code and supporting documentation are included where applicable.
The dataset can be reused to examine stakeholder-defined reforestation systems, compare perceptions of restoration constraints and enabling conditions, reproduce the network analysis and centrality rankings, test alternative network metrics, and support comparative research on restoration governance and stakeholder participation. Centrality values represent the structural prominence of factors within perceived causal networks and should not be interpreted as independently demonstrated causal effects.
The research was conducted with permission, including ethical approval for data collection, from Madagascar’s Ministry of Environment and Sustainable Development under permits 361/24/MEDD/SG/DGGE/DAPRNE/SCBE.Re and 206/225/MEDD/SG/DGGE/DAPRNE/SCBE.Re. All participants provided free, prior, and informed consent before participation. To protect participant confidentiality, the archived files exclude names, contact information, consent forms, audio recordings, photographs of participants, raw interview and focus-group transcripts, and other directly identifying information. Individual-level materials that could enable participant re-identification are not publicly shared. Users should appropriately cite the dataset and respect the ethical conditions under which the original information was collected.
Dataset DOI: 10.5061/dryad.pc866t266
Description of the data and file structure
Data from: Whose success counts? Reconciling stakeholder goals in Madagascar's reforestation policy
Associated research
Associated manuscript: Whose success counts? Reconciling stakeholder goals in Madagascar's reforestation policy
Target journal: Journal of Applied Ecology
Manuscript identifier: JAPPL-2026-00441
Dataset author and contact: Lanja Maminirina Rajaonarison
Data-collection period: 2024–2025
Study region: Madagascar
Study landscapes: Ambohitantely, Ankafobe, Ankarafantsika, and Andasibe
Dataset DOI: https://doi.org/10.5061/dryad.pc866t266
Data description
Dataset overview
This repository contains de-identified data products, documentation, and analysis outputs from a study using fuzzy cognitive mapping (FCM) to examine concepts and perceived causal relationships associated with reforestation success in Madagascar.
FCMs represent concepts as network nodes and perceived causal relationships as directed, weighted links. The analytical workflow converts individual FCM edge lists into adjacency matrices, checks matrix consistency, aligns matrices to a common set of harmonised concepts, calculates pooled and stakeholder-group mean matrices, and performs network analyses.
The analysis distinguishes four stakeholder categories:
Local_Community: local community participants or groups;
Leader_Community: community leaders;
Government: government and public-authority representatives;
NGO: non-governmental, conservation, technical, or project-based organisations.
These labels are analytical stakeholder categories, not measurement units.
The public repository contains de-identified derived analytical outputs. It does not contain participant names, contact details, interview recordings, verbatim transcripts, signed consent forms, direct identifiers, or a re-identification key.
Files and variables
Analytical units and network structure
The principal analytical units are:
Harmonised concepts or factors, represented as nodes in the cognitive networks.
Directed relationships between concepts, represented as weighted edges.
Individual FCMs, represented during analysis as directed adjacency matrices.
Pooled and stakeholder-group networks, created by averaging corresponding standardised matrices.
Network communities, identified using the Louvain community-detection algorithm.
A directed relationship from concept A to concept B does not imply an equivalent relationship from concept B to concept A.
Original edge-list variables used by the workflow
The individual input edge-list files used by the R workflow contain the following fields. These variables describe the analytical inputs; individual participant-level edge lists are not included among the public files listed in Section 2.
Source
Type: Character string.
Unit: Not applicable.
Definition: Concept from which a directed relationship originates.
Interpretation: Starting concept of the directed connection.
Cible
Type: Character string.
Unit: Not applicable.
Definition: Concept toward which the directed relationship points.
Interpretation: Target concept of the directed connection. Cible is the French term retained in the source data and R code.
Poids
Type: Numeric.
Unit: Original FCM elicitation scale.
Definition: Weight assigned to the directed relationship from Source to Cible.
Interpretation: The sign indicates the perceived direction of the relationship and the absolute value represents its recorded strength under the study's FCM elicitation protocol.
Important limitation: The R script alone does not define the permitted minimum and maximum of the original elicitation scale. Users must consult the study protocol and associated methodological documentation before interpreting the numerical range. The range must not be inferred from the variable name alone.
Each row represents one directed relationship. Before matrices are constructed, the R workflow converts concept labels to lower case, removes leading and trailing spaces, converts Poids to numeric, and excludes rows in which Source, Cible, or Poids is missing.
Interpretation of adjacency matrices
In an adjacency matrix:
the row identifies the source concept;
the column identifies the target concept;
the cell contains the directed relationship weight from the row concept to the column concept;
a positive value represents a relationship in the same direction under the study's FCM protocol;
a negative value represents a relationship in the opposite direction under the study's FCM protocol;
0 indicates that no non-zero relationship was entered for that ordered pair, or that the cell was filled with zero during alignment to the common concept set.
Standardisation aligns the concepts and dimensions of individual matrices. It does not rescale the original relationship weights. Pooled and stakeholder-group aggregate matrices are arithmetic means of the corresponding standardised matrices. Each included individual matrix has equal weight in the mean.
Files and variables
Public file inventory
The public dataset contains the following files:
Centrality_Metrics.csv
Centrality_Government.csv
Centrality_Leader_Community.csv
Centrality_Local_Community.csv
Centrality_NGO.csv
Top_10_Degree_Centrality.csv
Louvain_Communities.csv
Matrix_Consistency_Check.csv
R_Code.R
harmonised_factor.csv
The individual edge lists and participant-level matrices used as intermediate analytical inputs are not included in this public file inventory. The deposited CSV files are derived node-level results and quality-control outputs. Therefore, the public files support inspection of the reported analytical outputs and workflow, but do not by themselves permit complete reconstruction of every participant-level FCM from the original elicitation data.
Shared centrality-variable dictionary
The following variables occur in Centrality_Metrics.csv and in the four stakeholder-group centrality files.
Node
Type: Character string.
Unit: Not applicable.
Definition: Name of the harmonised concept represented by the network node.
Interpretation: Identifies the concept to which the calculated centrality values apply.
Label caution: Labels were converted to lower case and trimmed of leading and trailing spaces. Internal duplicate spaces, spelling variants, wording variants, and singular/plural forms may remain in archived outputs. Consult harmonised_factor.csv before combining, correcting, or renaming nodes.
Degree
Type: Numeric.
Unit: Dimensionless normalised network measure.
Calculation: igraph::degree(mode = "all", normalized = TRUE).
Interpretation: Normalised total degree considering incoming and outgoing connections. The calculation is based on the presence of connections, not on the numerical magnitude or sign of edge weights. Higher values indicate a larger number of direct connections relative to the size of the analysed graph.
Caution: A high degree indicates direct connectedness; it does not demonstrate causal importance, causal effect size, or intervention effectiveness.
Closeness
Type: Numeric; may be unavailable or undefined in some network configurations.
Unit: Dimensionless, algorithm-specific reciprocal-distance measure.
Calculation: igraph::closeness() with absolute edge weights supplied through the weights argument.
Interpretation: Describes how close a node is to other reachable nodes according to the weighted path distances used by the implemented algorithm. Higher values indicate shorter calculated path distances under this implementation.
Interpretation key: Values are not percentages and are not restricted to the interval from 0 to 1.
Methodological caution: In igraph, weights supplied to shortest-path functions are treated as distances or lengths. Because the workflow supplies absolute FCM weights directly, larger absolute relationship weights are treated as longer distances rather than being transformed from relationship strength to distance. The reported values must therefore be interpreted according to this implementation and should not be treated as conventional strength-to-distance transformed closeness centrality.
Missing values: NA indicates that the metric was unavailable or undefined under the implemented calculation. It does not mean that the node was absent from the source network.
Betweenness
Type: Numeric.
Unit: Dimensionless normalised network measure.
Calculation: igraph::betweenness() with absolute edge weights supplied through the weights argument.
Interpretation: Measures the extent to which a node lies on calculated shortest paths between other nodes.
Methodological caution: The supplied absolute weights are treated as distances or lengths. Larger absolute FCM relationship weights are therefore treated as longer paths in the implemented calculation. The values should be interpreted cautiously and not as direct measures of causal control or empirical mediation.
Zero: A value of 0 is a calculated result indicating that the node does not lie on any of the relevant shortest paths under the implemented calculation; it is not missing data.
Eigenvector
Type: Numeric.
Unit: Dimensionless, algorithm-specific centrality score.
Calculation: igraph::eigen_centrality(directed = TRUE, weights = ...).
Interpretation: A node receives a higher score when it is connected to other structurally influential nodes, taking the absolute edge weights used by the workflow into account. Values are relative to the analysed network and are not percentages.
Katz
Type: Numeric.
Unit: Dimensionless scaled authority score.
Calculation: Despite the archived column name, the R code calculates this variable using igraph::authority_score(scale = TRUE, weights = ...).
Interpretation: Higher values indicate stronger authority-type centrality under the implemented calculation.
Important naming caution: This variable is not Katz centrality, despite being labelled Katz in the archived output. The original column name is retained to remain consistent with the deposited files and the existing analytical outputs. Any reuse must describe this variable as a scaled authority score calculated with authority_score().
Before Closeness, Betweenness, Eigenvector, and the Katz-labelled authority score are calculated, the workflow replaces edge weights with their absolute values. Positive and negative relationships of equal magnitude therefore contribute equally to these calculations. Relationship signs remain represented in the underlying aggregated adjacency matrices but are not retained in the reported centrality scores.
Centrality_Metrics.csv
Description: Centrality measures for the pooled analytical network.
Unit of observation: One harmonised concept or network node.
Variables: Node, Degree, Closeness, Betweenness, Eigenvector, and Katz, as defined in Section 2.2.
Missing-value note: Some Closeness values may be NA where the metric was unavailable or undefined under the implemented calculation.
Centrality_Government.csv
Description: Centrality measures for the aggregated network representing government and public-authority stakeholders.
Unit of observation: One retained harmonised concept or network node.
Variables: Node, Degree, Closeness, Betweenness, Eigenvector, and Katz, as defined in Section 2.2.
Missing-value handling: No missing values are present in the deposited file. According to the archived workflow, rows containing NA in any calculated centrality field were removed before the stakeholder-group CSV was written.
Zero values: Calculated zeros are valid analytical results, not missing data.
Centrality_Leader_Community.csv
Description: Centrality measures for the aggregated network representing community leaders.
Unit of observation: One retained harmonised concept or network node.
Variables: Node, Degree, Closeness, Betweenness, Eigenvector, and Katz, as defined in Section 2.2.
Missing-value handling: No missing values are present in the deposited file. According to the archived workflow, rows containing NA in any calculated centrality field were removed before the stakeholder-group CSV was written.
Zero values: Calculated zeros are valid analytical results, not missing data.
Centrality_Local_Community.csv
Description: Centrality measures for the aggregated network representing local community participants or groups.
Unit of observation: One retained harmonised concept or network node.
Variables: Node, Degree, Closeness, Betweenness, Eigenvector, and Katz, as defined in Section 2.2.
Missing-value handling: No missing values are present in the deposited file. According to the archived workflow, rows containing NA in any calculated centrality field were removed before the stakeholder-group CSV was written.
Zero values: Calculated zeros are valid analytical results, not missing data.
Label caution: Some archived labels may retain internal spacing, spelling, wording, or singular/plural variants. Users must consult harmonised_factor.csv before merging or renaming apparently related concepts.
Centrality_NGO.csv
Description: Centrality measures for the aggregated network representing non-governmental, conservation, technical, or project-based organisations.
Unit of observation: One retained harmonised concept or network node.
Variables: Node, Degree, Closeness, Betweenness, Eigenvector, and Katz, as defined in Section 2.2.
Missing-value handling: No missing values are present in the deposited file. According to the archived workflow, rows containing NA in any calculated centrality field were removed before the stakeholder-group CSV was written.
Zero values: Calculated zeros are valid analytical results, not missing data.
Top_10_Degree_Centrality.csv
Description: Reporting subset containing the ten concepts with the highest Degree values after excluding the focal concepts ankarafantsika reforestation and reforestation.
Unit of observation: One selected harmonised concept.
Ordering: Rows are ordered from higher to lower Degree.
Variables:
Node: Harmonised concept name; character string; no unit.
Degree: Normalised total degree copied from Centrality_Metrics.csv; numeric and dimensionless; the variable used for ranking.
Closeness: Closeness value copied from Centrality_Metrics.csv; numeric and dimensionless.
Betweenness: Betweenness value copied from Centrality_Metrics.csv; numeric and dimensionless.
Eigenvector: Eigenvector-centrality value copied from Centrality_Metrics.csv; numeric and dimensionless.
Katz: Katz-labelled scaled authority score copied from Centrality_Metrics.csv; numeric and dimensionless. This is not Katz centrality.
This file is a reporting subset and must not be used as a substitute for the complete pooled centrality file.
Louvain_Communities.csv
Description: Community membership for each node in the pooled network.
Unit of observation: One harmonised concept.
Node
Type: Character string.
Unit: Not applicable.
Definition: Concept represented by a node in the pooled network.
Community
Type: Integer categorical identifier.
Unit: Not applicable.
Definition: Community-membership number assigned by the Louvain algorithm.
Interpretation: Nodes with the same number belong to the same detected network community.
Interpretation key: Community numbers are arbitrary category labels. They are not ranks, effect sizes, probabilities, or measures of community quality.
For community detection, the directed pooled network is converted to an undirected network using as.undirected(mode = "collapse", edge.attr.comb = "sum"). Louvain membership is therefore based on the resulting undirected, collapsed network and its combined edge weights, not directly on the original directed network.
Matrix_Consistency_Check.csv
Description: Quality-control report comparing the concept labels represented in the rows and columns of each raw adjacency matrix.
Unit of observation: One checked adjacency-matrix file.
File
Type: Character string.
Unit: Not applicable.
Definition: Name of the raw adjacency-matrix file examined.
Interpretation: Identifies the matrix to which the quality-control result applies. Filenames are analytical source identifiers and must not be used to infer participant identity.
Missing_in_Rows
Type: Character string containing comma-separated concept labels, or an empty field.
Unit: Not applicable.
Definition: Concept labels present among the matrix column headers but absent from the row labels.
Interpretation key: An empty field means that no concept was missing from the rows. A populated field lists unmatched column labels.
Missing_in_Cols
Type: Character string containing comma-separated concept labels, or an empty field.
Unit: Not applicable.
Definition: Concept labels present among the matrix row labels but absent from the column headers.
Interpretation key: An empty field means that no concept was missing from the columns. A populated field lists unmatched row labels.
Missing_in_RowsandMissing_in_Cols do not describe missing participants, missing interviews, or missing numerical observations. They describe mismatches between the row and column concept-name sets. In the deposited report, an empty field means no inconsistency detected, not unknown or unavailable data.
R_Code.R
Description: R source code implementing the FCM matrix-processing and network-analysis workflow.
The script:
reads individual FCM edge lists;
cleans concept labels and converts relationship weights to numeric;
creates directed adjacency matrices;
checks row-column matrix consistency;
aligns individual matrices to a common concept set;
calculates pooled and stakeholder-group arithmetic mean matrices;
creates directed graphs and calculates centrality measures;
identifies high-degree concepts;
detects Louvain communities;
generates comparative figures and multivariate analyses.
The archived script contains local Windows directory paths. Anyone rerunning the workflow must replace these paths and reproduce the expected input and output folder structure. The public repository does not contain all participant-level inputs used by the complete workflow.
harmonised_factor.csv
Description: A two-column reference table documenting the harmonisation of factor or concept labels used in the FCMs and network analyses.
File format: Comma-separated values (CSV), suitable for inspection and reuse in spreadsheet software and statistical/programming environments.
Rows and columns: The file contains 450 data rows and two columns. The first column, Simplified Factor, contains the harmonised factor labels. The second column, Raw factors mentioned in FCM, contains the original factor expressions recorded in the FCMs and mapped to the corresponding harmonised labels.
Interpretation of blank cells: A non-empty cell in Simplified Factor introduces a harmonised factor. Subsequent rows with a blank Simplified Factor cell list additional raw expressions associated with the most recent non-empty harmonised factor above. These blank cells are therefore part of the table's grouping structure and do not indicate missing or unknown factor information.
Purpose: Shows how different expressions recorded during FCM data collection were grouped under harmonised analytical labels for matrix alignment and network analysis.
Use requirement: Consult this file before renaming, combining, correcting, or interpreting apparently similar Node labels. When importing the CSV into software, retain the row order and use the most recent non-empty Simplified Factor value to identify the harmonised group for each subsequent raw-factor entry until the next non-empty harmonised factor appears.
The file is a factor-harmonisation reference table; it is not a participant-level dataset and does not contain participant identifiers.
Relationships among deposited files
The workflow and deposited outputs are related as follows:
Individual FCM edge lists were converted into adjacency matrices by R_Code.R.
Matrix_Consistency_Check.csv records whether each checked matrix had matching row and column concept sets.
Factor labels were standardised with reference to the harmonisation documented in harmonised_factor.csv.
Individual standardised matrices were averaged to create pooled and stakeholder-group networks.
Centrality_Metrics.csv contains node-level measures for the pooled network.
The four stakeholder-group centrality files contain corresponding measures for each stakeholder category.
Top_10_Degree_Centrality.csv is a selected reporting subset derived from the pooled centrality results.
Louvain_Communities.csv contains community membership for the pooled network.
R_Code.R documents the computational workflow used to create and analyse these outputs.
The individual edge lists, participant-level matrices, and aggregated adjacency matrices are not included in the public file inventory described here. Users should therefore distinguish between workflow steps documented by the code and files actually available in the public repository.
Missing values, zeros, and data handling
The following cases must be distinguished:
Missing input fields: Rows with missing Source, Cible, or Poids are removed before an adjacency matrix is created.
Zero in an adjacency matrix: No non-zero relationship was recorded for that ordered pair, or the cell was filled with zero during alignment. It is not automatically equivalent to unknown data.
Zero in a centrality file: A calculated zero; it is not missing data.
Empty Missing_in_Rows or Missing_in_Cols field: No row-column label inconsistency was detected.
Populated Missing_in_Rows or Missing_in_Cols field: The field lists concept labels that did not occur in both matrix dimensions.
NAinCentrality_Metrics.csv: The relevant metric was unavailable or undefined for that node under the implemented calculation.
NAduring stakeholder-group centrality calculation: According to the archived workflow, rows with anNA in any centrality field are removed before the stakeholder-group CSV is written.
Missing values during PCA preparation: The R workflow replaces NA values in vectorised matrix data with zero before scaling and PCA.
Interpretation guidance and methodological cautions
Centrality does not demonstrate causality
Centrality measures describe structural positions in the analysed cognitive networks. They do not estimate empirical causal effects, intervention effectiveness, statistical mediation, or the probability that an intervention will succeed.
Relationship direction and absolute weights
The original adjacency matrices retain signed directed relationships. However, the centrality workflow uses abs(weight). Positive and negative relationships of equal magnitude therefore contribute equally to the reported weighted centrality calculations.
Degree
Degree measures direct connectedness based on incoming and outgoing links. A high value indicates many direct connections relative to graph size. It does not by itself demonstrate causal importance.
Closeness and betweenness
The workflow passes absolute FCM weights directly to igraph shortest-path functions, where weights are interpreted as distances or lengths. Consequently, stronger absolute relationships are treated as longer rather than shorter paths. The reported closeness and betweenness values reflect this specific implementation and should be interpreted cautiously.
Eigenvector centrality
Eigenvector measures recursive structural prominence. A concept receives a higher score when it is connected to other highly central concepts. Values are relative to the analysed network.
Katz-labelled authority score
The variable named Katz is calculated using igraph::authority_score(). It must therefore be interpreted as a scaled authority score, not as Katz centrality. The archived name is retained only for consistency with the deposited outputs.
Louvain communities
Louvain communities are algorithmic network partitions, not predefined thematic groups or stakeholder categories. Community identifiers are arbitrary and may change if network construction, edge treatment, software versions, or algorithm settings change.
Factor labels
Concept labels are converted to lower case and trimmed of leading and trailing spaces. Other spelling differences, internal duplicate spaces, wording variants, and singular/plural forms are not necessarily reconciled automatically. Users must consult harmonised_factor.pdf and the study protocol before merging or renaming nodes.
Code/software
Code and software
The analysis was written in R and uses the following packages:
readxl
dplyr
tidyr
stringr
scales
tibble
igraph
ggraph
ggplot2
fmsb
readr
forcats
ggtext
ggfortify
dendextend
The archived script does not record the R version or package versions. Exact reproduction may therefore depend on the software environment. Users should document the R version, package versions, changes to paths, and any modifications made before rerunning any part of the workflow.
Access information
Human-participant data and de-identification
The dataset is derived from research involving human participants. The public repository contains de-identified derived network-analysis outputs rather than identifiable raw qualitative materials.
The public files exclude:
participant names and contact details;
interview recordings;
verbatim transcripts;
signed consent forms;
direct personal identifiers;
a re-identification key;
raw narrative material that could directly reveal participant identity.
Stakeholder information is represented through analytical categories, concept labels, derived network measures, and source-matrix identifiers. Users must not attempt to re-identify participants from roles, organisations, study locations, filenames, or combinations of contextual information. Any reuse must preserve participant confidentiality.
Individual participant-level FCM edge lists and matrices are not included in the public file inventory. The public deposit is limited to de-identified derived analytical outputs, documentation, code, and quality-control results.
Access and reuse information
The dataset is intended to support inspection of the associated analysis and secondary research on reforestation governance, stakeholder perspectives, fuzzy cognitive mapping, and network structure.
When reusing these files:
Cite both the Dryad dataset and the associated article.
Retain original labels when reproducing the archived analysis.
Consult harmonised_factor.pdf before combining, correcting, or renaming factors.
Preserve the distinction between calculated zeros, empty consistency-check fields, and missing or undefined values.
Do not interpret centrality scores as empirical causal-effect estimates.
Do not describe the Katz column as Katz centrality; identify it as a scaled authority score calculated with authority_score().
Account for the direct use of absolute FCM weights as shortest-path distances when interpreting Closeness and Betweenness.
Do not attempt to identify research participants.
Document any cleaning, relabelling, thresholding, transformation, path modification, or software change.
Do not claim complete participant-level reproduction from the public files because the individual FCM inputs are not included in the public inventory.
During peer review, access may be restricted to authorised editors and reviewers through Dryad's private-review process. Following curation and publication, the dataset DOI and public-access information will be provided by Dryad, subject to the repository's applicable terms and any approved access arrangement.
Known limitations
The public repository contains de-identified derived analytical outputs rather than unrestricted raw interview materials or participant-level FCM inputs.
Complete reconstruction of every participant-level FCM is not possible from the public files alone.
Network results depend on factor harmonisation, matrix aggregation, edge treatment, and software settings.
The original FCM weight range is not defined by the R script alone and must be checked against the study protocol.
Relationship signs are not retained in the centrality calculations because absolute edge weights are used.
ClosenessandBetweenness use absolute FCM weights as path distances without a strength-to-distance transformation.
The output labelled Katz is an authority score rather than Katz centrality.
Some pooled closeness values may be unavailable or undefined.
Community assignments are algorithm-dependent.
Internal spacing, spelling, wording, or singular/plural variants may remain in archived node labels and should not be merged without checking harmonised_factor.pdf.
The R version and package versions are not recorded in the archived script.
The archived R script contains local file paths that must be changed before reuse.
Human subjects data
This dataset is derived from research involving human participants conducted in Madagascar during 2024–2025. Explicit informed consent was obtained from all participants, including consent for the publication and sharing of de-identified research data in the public domain.
The publicly available dataset contains only de-identified and derived analytical outputs. Participant names, contact details, and other direct personal identifiers have been excluded. Interview recordings, verbatim transcripts, signed consent forms, and any re-identification keys are not included in the repository.
Individual responses have been represented through harmonised concepts, stakeholder categories, aggregated adjacency matrices, and derived network-analysis results. No direct personal identifiers are included in the deposited files. These measures were taken to protect participants' confidentiality and reduce the risk of identification through contextual information.
The publicly available files do not contain original individual-level interview materials or participant-level FCM inputs.
Data were collected between 2024 and 2025 across four protected-area landscapes in Madagascar: Ambohitantely, Ankafobe, Ankarafantsika and Andasibe. Participants were purposively recruited based on their knowledge and experience of reforestation or tree-planting. The study combined semi-structured interviews, focus-group discussions, and fuzzy cognitive mapping. Sessions were conducted in Malagasy and, where appropriate, French or English.
During the fuzzy cognitive mapping exercises, participants identified factors perceived to influence reforestation success or failure and described the causal relationships among these factors. Relationships were assigned a positive or negative direction and a strength from 1, weak, to 4, very strong. Combining direction and strength produced signed edge weights ranging from −4 to +4. Completed maps were transcribed into weighted adjacency matrices, with rows representing source factors and columns representing receiving factors. A value of zero indicates that no causal relationship was recorded.
Factor labels were reviewed and harmonised to allow comparison across maps, stakeholder groups and landscapes. Synonymous or closely related expressions were grouped under common factor labels, while opposing or conceptually distinct factors were retained separately. This procedure reduced 442 raw factor expressions to 191 harmonised factors.
Following factor harmonisation, individual adjacency matrices were aggregated within four stakeholder groups: government and local authorities, NGOs and technical organisations, community leaders, and local communities. Each fuzzy cognitive map was given equal weight, and signed edge values were averaged within stakeholder groups. Network analysis was conducted in R using the igraph package. Katz centrality was calculated using igraph::alpha_centrality() to identify factors occupying structurally prominent positions within each stakeholder network. Factors were ranked independently within each stakeholder group.
Before archiving, the files were reviewed to remove direct identifiers and information presenting a material risk of participant re-identification. Raw transcripts, recordings, consent documentation, and personally identifiable information are not included in the archived dataset.
