A segmented landowner dataset for a mountain protected-area gateway: Cerro Castillo National Park, Chilean Patagonia (2024)
Description
This dataset supports a landowner-typology study of the buffer zone of Cerro Castillo National Park, a mountain protected-area gateway in the Aysén region of Chilean Patagonia undergoing tourism-driven amenity transition. It contains de-identified responses from a survey of 641 landowners conducted Feb–July 2024 within 10 km of the park (637 retained for segmentation; 636 georeferenced to watershed sub-catchments), with the derived variables, segment assignments and validation results needed to reproduce the analysis. Four owner segments — Committed newcomers, Rooted landholders, Active citizens, and Park-disengaged owners — are obtained by k-means on eleven standardised variables: two territorial-bond indices (bond with own land, bond with the park), a disposition-to-change index, four conservation-citizenship sub-indices (civic identity, civic behaviour, norm compliance, participatory action), age, education, tenure length and acquisition mode. Because the bond items were administered under two survey versions (A/B), the two bond indices are harmonised across versions by mean-sigma random-groups test-equating before clustering, removing the instrument artifact; the equated inputs (idx_tierra_h, idx_parque_h) are stored alongside the raw indices. The workbook has seven sheets: (1) de-identified case-level data (641 × 191) — segment labels, all clustering inputs, the equated bond indices, an assignment-method flag, and profiling, spatial and institutional variables; (2) a codebook with variable descriptions, types and value labels (Spanish glossed in English); (3) Derived_variables — the construction of every derived index, verified to reproduce the stored values; (4) Validation — internal indices (silhouette, Davies–Bouldin) for k = 2–8 and nonparametric bootstrap stability; (5) a segment-by-sticker crosstab from a qualitative sub-sample; (6) a sub-catchment table of person- versus land-area dominance; and (7) a README documentation sheet. No direct or indirect identifiers are included; geography is retained only as coarse watershed sub-catchment (1–19). Data were collected under FONDECYT Regular project 1230020 (ANID, Chile); ethics approval IRI11_23, Universidad Austral de Chile; participants gave informed consent. This is the segmentation/typology dataset, distinct from the companion deposit (Gale & Báez-Montenegro, 2026; Mendeley Data, doi:10.17632/cf69gj7ndg.1), which holds the relational-values and place-attachment survey underlying a separate structural-equation analysis. The two share the 2024 field survey but differ in scope and structure: the present file provides the segment assignments, the equated (harmonised) clustering inputs, an education variable absent from the companion deposit, spatial and institutional variables, validation material, and the item batteries needed to reconstruct and verify every derived index; the companion deposit organises the relational-values and place-attachment items for structural-equation modelling.
Files
Steps to reproduce
Data collection. Survey property owners in the inhabited sub-catchments of the Cerro Castillo NP buffer zone (<10 km), Feb–Jul 2024. Items cover property, land use, future intentions, citizenship (Q29) and disposition-to-change (Q28) batteries, sub-catchment, org. participation, socio-demographics, a sticker choice, and an A/B psychographic block: version A (n=310, relational values) and B (n=331, place attachment), toward parcel and park. Ethics IRI11_23 (Universidad Austral de Chile); informed consent. De-identification. Remove direct/indirect identifiers (names, email, IP, GPS, response IDs, timestamps, free-text); anonymise IDs; blank missing placeholders; keep geography only as sub-catchment (1–19). Derived indices (0-100, item means). idx_tierra/idx_parque from own-land and park batteries (Q23/Q24 ver. A; Q26/Q27 ver. B). From Q29: cit_norm_compliance, cit_civic_behaviour, cit_do_democracy, cit_civic_identity. disposition_change from Q28. Recodes: age=Q35 midpoint; education=ordinal Q36; acquisition=inherited vs not. Formulas in the Codebook sheet. Harmonisation. Equate idx_tierra and idx_parque across versions A/B by mean-sigma random-groups test-equating (Kolen & Brennan, 2014) → idx_tierra_h, idx_parque_h. Segmentation. Standardise the eleven inputs (two equated bonds, disposition_change, four cit_*, age, education, tenure, acquisition); k-means k=4, nstart=100, iter.max=1000, seed=42 on the 635 complete cases; assign 2 owners missing only education by nearest centroid (637 segmented). Profile on held-out variables (Kruskal–Wallis; chi-square). Validation. Silhouette and Davies–Bouldin, k=2–8; bootstrap stability (Hennig, 2007; B=500; mean Jaccard). Spatial. Estimate parcel area from size-band midpoints (top band 600 ha); per sub-catchment, plurality segment by owner count vs by area (divergence = flip). Software. R (stats::kmeans; cluster; nnet; MASS; optional readxl/psych). Run CCNP_landowner_segmentation_REPLICATION.R; seed 42 reproduces it.