Data and R code for a Data Envelopment Analysis of national innovation system efficiency in the EU-27, 2022–2024
Description
This record contains the compiled dataset and R code for a Data Envelopment Analysis of national innovation system efficiency across the 27 EU member states. Each country is treated as a unit converting research inputs into scientific and technological outputs, and efficiency is estimated relative to the observed best-practice frontier. All indicators are three-year means for 2022–2024, expressed in natural units. Raw extracts are drawn from Eurostat, the OECD Bibliometric Indicators database, the European Innovation Scoreboard 2025, and the WIPO Global Innovation Index 2025. The data/ folder holds these source extracts, the intermediate workbooks used to assemble them, and two analysis-ready files: NIS_DEA_data.csv, containing the six candidate inputs and three outputs used in the specification, and NIS_SII_2025.csv, containing the Summary Innovation Index scores used for external comparison. All nine candidate variables that reached the specification stage are retained in NIS_DEA_data.csv, including the two inputs subsequently dropped by the correlation screen, so the screen itself can be reproduced. A raw extract for employment in knowledge-intensive activities, excluded earlier on conceptual grounds, is also included. The R scripts/ folder contains nine numbered scripts covering the main DEA estimation, the Simar–Wilson bootstrap, the rank comparison with the Summary Innovation Index, returns-to-scale classification, collinearity screening, an IP-revenue sensitivity check, external validity against the Global Innovation Index, and a stochastic frontier identifiability test. Run them in numerical order with the working directory set to the pack root. Outputs are written to results/, which already contains the files from a complete run, including a session log. Estimation used R 4.6.1 with the Benchmarking package version 0.33; the bootstrap seed is fixed at 123. Source extracts in data/ are redistributed from Eurostat, the OECD, the European Innovation Scoreboard and WIPO, and remain subject to the terms of their respective publishers. The CC BY 4.0 licence applies to the compiled dataset, derived files, R code and results produced by the authors.
Files
Steps to reproduce
Software. R 4.6.1 or later, with packages Benchmarking (0.33) and readxl (1.5.0). Benchmarking loads lpSolveAPI, ucminf and quadprog as dependencies. Setup. Download the record and preserve the folder structure: data/, results/ and R scripts/ must sit as siblings under one root. Set the R working directory to that root, not to the script folder — for example setwd("/path/to/pack"). All paths in the scripts are relative to it. Run order. Run 01_main_analysis.R first: it produces TableA_VRS_CRS_scale_supereff.csv and TableB_bootstrap_biascorrected.csv in results/, and 02_rq1_correlation.R reads Table B. The remaining scripts read only from data/ and can be run in any order, though numerical order is the logical sequence. 06_gii_externalvalidity.R takes the bias-corrected scores as a hard-coded vector reproducing Table B, so it runs independently. Reproducibility. The Simar–Wilson bootstrap in script 01 uses 2,000 replications with the random seed fixed at 123, so bias-corrected scores and confidence intervals reproduce exactly. All other procedures are deterministic. Expected results. Six countries on the constant-returns frontier (Denmark, Cyprus, Luxembourg, Malta, Netherlands, Finland). Bias-corrected scores from 0.7693 for Italy to 0.1281 for Bulgaria. Four super-efficiency infeasibilities under variable returns. Returns to scale: 6 optimal, 20 increasing, 1 decreasing. Spearman correlation with the Summary Innovation Index of 0.808 bias-corrected and 0.807 raw, N = 27. Correlation with the Global Innovation Index efficiency ratio of 0.524, p = 0.00565. Full and reduced rankings in the IP-revenue sensitivity check correlate at 0.998. results/ already contains the output of a complete run, including the session log, so a re-user can compare without executing anything.
Institutions
- Budapest University of Technology and EconomicsBudapest, Budapest