Replication Package: From Breaks to Bridges

Paper: "From Breaks to Bridges: An Architecture for Survey Redesigns That Expand Conceptual Scope"

Authors: Germán Tessmer (Universidad Nacional de Rosario) and Bárbara Boggiano (Universidad Alberto Hurtado)

Contact: german.tessmer@unr.edu.ar | Project page

---

Overview

This repository contains the complete R pipeline to reproduce all results, figures, and tables in the paper. The pipeline constructs a bridged labour-formality series for Argentina (2016Q4--2025Q3) from EPH microdata, combining deterministic harmonization, a two-factor latent structure, LASSO-selected predictive models (LPM, GLM, SLS), and a hybrid backcasting rule.

The paper's online appendices are included in this repository at appendix/2026_FromBreakstoBridges-OnlineAppendix.pdf.

Requirements

• R >= 4.5.0
• renv for package management (run renv::restore() to install all dependencies from renv.lock)
• Hardware: 16 GB RAM recommended; 7+ CPU cores for parallel bootstrap and LOCO cross-validation
• OS: Developed on Windows 11; should work on Linux/macOS with minor path adjustments in parametros.R

Data Sources

1. EPH Microdata (primary): Pre-processed panel files from the EPH--Observatorio repository. Please cite this dataset as:

   Tessmer, Germán; Moleres, Santiago; Hess, Laureano Elian, 2024, "Encuesta Permanente de Hogares de Argentina (EPH - Observatorio). Explicada, etiquetada y ampliada", https://doi.org/10.57715/UNR/BL85Z8, RDA UNR, V3.

   • DOI: 10.57715/UNR/BL85Z8
   • Place the .RData files in the directory specified by RUTA_BASES in script/config/parametros.R (default: C:/oes/eph_rdos/capa2/).

2. SIPA Administrative Data: Published by the Secretaría de Trabajo, Argentina.

   • URL: https://www.argentina.gob.ar/trabajo/estadisticas/situacion-y-evolucion-del-trabajo-registrado
   • Used only for external validation (Layer 7, script 10e).

3. INDEC Official Microdata: Original EPH individual and household databases.

   • URL: https://www.indec.gob.ar/indec/web/Institucional-Indec-BasesDeDatos

No restricted-access or confidential data are used.

How to Run

1. Clone this repository and open the R project.
2. Install dependencies: renv::restore()
3. Edit script/config/parametros.R:
   • Set RUTA_BASES to the directory containing the EPH .RData files.
   • Adjust N_CORES if needed (default: auto-detected minus 1).
4. Run the full pipeline: source("00_master_runner.R")

The master runner executes all layers sequentially. Total runtime is approximately 4--6 hours depending on hardware (the heterofactor estimation in Layer 3 is the bottleneck).

Individual layers can also be run independently by sourcing scripts in order.

Pipeline Architecture (7 Layers)

Layer	Directory	Description	Key Scripts
0	script/00_diccionarios/	Variable dictionaries and crosswalks	00_diccionarios.R
1	script/01_datos_base/	Panel construction, taxonomy, income cleaning, ICH	01 -- 03c
2	script/02_proxies/	Proxy system for latent factors	04, 05
3	script/03_heterofactor/	Two-factor latent model (FIML)	06a -- 06d
4	script/04_modelado/{LPM,GLM,SLS}/	LASSO model estimation (3 families)	07a -- 07e
5	script/05_backcasting/{LPM,GLM,SLS}/	Historical backcasting and hybrid construction	08a -- 08c
6	script/06_comparativo/	Cross-model comparison	09a -- 09c
7	script/07_robusto/	Robustness checks (16 scripts)	10a -- 10o

Configuration

All paths, seeds, temporal parameters, and model suffixes are defined in script/config/parametros.R. Key settings:

• ANIO_INI / TRIM_INI -- ANIO_FIN / TRIM_FIN: Panel time range (default: 2016Q4--2025Q3)
• N_TRIMESTRES_TRAINING: Number of overlap quarters used for model training (default: 4)
• SEED_GLOBAL: Random seed for reproducibility (default: 123)
• N_CORES: Parallel workers for bootstrap and cross-validation

Outputs

All outputs are written to rdos/ (created automatically):

• rdos/datos/ -- Processed panel files (.rds)
• rdos/modelos/ -- Fitted models (.rds)
• rdos/contratos/ -- Validation contracts with citable statistics (.rds)
• rdos/reportes/ -- HTML diagnostic reports and CSV summaries
• rdos/figuras/ -- All figures as PDF files

Language Note

The R scripts are written in Spanish (comments, variable names, diagnostic messages). Each script includes a bilingual header summarizing its purpose, inputs, and outputs in English. The pipeline structure and variable naming follow a consistent convention documented in the headers and in the variable dictionary (rdos/inputs/diccionarios/).

License

MIT License. See LICENSE for details.