Dataset, Survey Instruments, and R Script for "Agentic Orchestration in Geospatial Programming: AI Agent Effectiveness as a Function of Spatial Complexity and User Experience in a Latin American Graduate Context"

This repository contains the complete anonymized dataset, the four ad hoc survey instruments in XLSForm format, and the R analysis script supporting the research article "Agentic Orchestration in Geospatial Programming: AI Agent Effectiveness as a Function of Spatial Complexity and User Experience in a Latin American Graduate Context," submitted to *ISPRS International Journal of Geo-Information*. The study evaluated the use and effectiveness of AI agents (Google Antigravity) as support tools for orchestrating intelligent solutions to automate geospatial workflows of increasing complexity, within the Geospatial Programming: Tools for Innovation module of the Geoinformation Management Specialization at Universidad Técnica Particular de Loja (UTPL), Ecuador. Data were collected through four longitudinal surveys administered at strategic moments of the module (baseline, post-Workshop 1 vector, post-Workshop 2 raster, and module closing) to a cohort of 20 graduate students, of whom 15 met all quality control criteria and constituted the final analytical sample. The analysis examines six hypotheses (H1a, H1b, H2, H3, H4, H5) relating prior programming and GIS experience to prompt quality, iteration efficiency, perceived agent usefulness across spatial complexity levels, competence evolution, and qualitative perceptions of the human-agent interaction. The surveys were designed ad hoc for this specific study and were not subjected to formal psychometric validation; they are provided here to facilitate replication and adaptation in other geoinformatics training contexts. Contents 1. Respuestas_Enc0.xlsx — Survey 0 (Baseline, week 1): Raw responses containing demographic data (Section A: undergraduate training area, employment sector, type of prior Python experience), prior experience in programming and GIS (Section B: 10 items on 0-10 scales covering general programming, Python, five GIS dimensions, AI tools, geospatial competence, and orchestration capacity), and attention/honesty control items (C1, C2). Collected before any exposure to the agentic platform. All responses in Spanish. 2. Respuestas_Enc1.xlsx — Survey 1 (Post-Workshop 1 — Vector, week 5): Raw responses containing support resource use and perceived usefulness for four pedagogical resources (Section D: 11 items including iteration count D9 and AI model selection D10), agent orchestration quality metrics including prompt precision and output fidelity (Section E: 4 items), and vector result validation with error detection categories (Section F: 5 items). Includes honesty control C3. All responses in Spanish. 3. Respuestas_Enc2.xlsx — Survey 2 (Post-Workshop 2 — Raster, week 7): Raw responses containing resource use in the raster context (Section G: 11 items, parallel structure to Section D), agent orchestration quality for NDWI spectral index tasks including study area specification and technical prompt elements (Section H: 4 items), and raster result validation with domain-specific error categories (Section I: 6 items). Includes honesty control C5. All responses in Spanish. 4. Respuestas_Enc3.xlsx — Survey 3 (Module Closing, week 8): Raw responses containing competence evolution measures (Section J: 4 items including post-module geospatial competence J1, orchestration capacity J2, and perceived role J4), global retrospective agent perception across three spatial complexity levels (Section K: 6 items including usefulness for non-spatial K1, vector K2, and raster K3 tasks), and professional projections including concern scales and free-text recommendations (Section L: 7 items). Includes attention control C7 and calibration control C8. All responses in Spanish. 5. analisis_EUEA_PG.R — Comprehensive R analysis script implementing the full predefined statistical analysis plan. The script performs: - Data import and longitudinal integration of the four survey files via successive left joins by student ID - Data type coercion and cleaning with attention/honesty control filtering (C2 = 4, C3 = 4, C5 = 5, C7 = 3) - Shapiro-Wilk normality testing for all key variables - Principal Component Analysis (PCA) on GIS experience variables (B3-B7) with KMO (0.542) and Bartlett sphericity test (χ² = 34.56, p < 0.001) prerequisites - Bivariate correlations (Pearson and Spearman) for hypothesis testing (H1a: B2→D9/G9; H1b: GIS index→E1/H1; H3: E1→D9, E4→D9, E1→errors, H1→G9, H1→errors) - Friedman test with post-hoc pairwise Wilcoxon signed-rank tests and Bonferroni correction for perceived usefulness across complexity levels (H2: K1, K2, K3) - Wilcoxon signed-rank tests for paired pre-post competence comparisons (H4: B9→J1, B10→J2) - NRC sentiment analysis on free-text recommendations L5 (H5) - Ordinal logistic regression via MASS::polr for iteration prediction - Generation of publication-ready bilingual (Spanish/English) figures (PNG + SVG) and Excel tables (T1-T5) - Requires R >= 4.4.3 and the following packages: readxl, dplyr, tidyr, ggplot2, ggpubr, corrplot, scales, writexl, svglite, broom, tibble, MASS, factoextra, psych, syuzhet 6. EUEA_PG_Enc0.xlsx, EUEA_PG_Enc1.xlsx, EUEA_PG_Enc2.xlsx, EUEA_PG_Enc3.xlsx — Survey instruments in XLSForm format (sheets: survey, choices, settings) used to configure the ArcGIS Survey123 forms. These define the item wording, response options, skip logic, and relevance conditions for all 81 items (73 thematic + 8 control). Provided for replication and adaptation purposes. Study Purpose To evaluate the use and effectiveness of AI agents as support tools in the orchestration of intelligent solutions for automating geospatial workflows of increasing complexity, analyzing how graduate students guide the agent through technical instructions and validate the generated results as a function of their prior programming and GIS experience. Recommended Use - Replication of the statistical analyses reported in the article - Adaptation of the survey instruments for studies on AI agent use in other programming or geoinformatics training contexts - Secondary analysis of AI agent adoption patterns in geospatial education - Teaching applications in courses on research methods, nonparametric statistics, or AI-assisted programming pedagogy File Format Excel (.xlsx) for survey response data and XLSForm instruments; R script (.R) for statistical analysis. Language Spanish (survey responses, item labels, and XLSForm instruments); English and Spanish (analysis script output, bilingual figures and tables).