2026-10-13T06:36:51Z https://data.mendeley.com/oai

oai:data.mendeley.com/m3jj8n89c8.1 2026-06-12T02:45:20Z

KIM, SANGSOON Convergence masking and the innovation paradox in OECD panels Mendeley Data This repository contains the reproduction package for the paper "Convergence masking and the innovation paradox in OECD panels" (S. Kim and Y.-W. Sawng). Research hypothesis. In two-way fixed-effects panel regressions of technological change on R&D intensity, omitting the lagged technology level biases the R&D coefficient toward zero. Proximity to the technology frontier raises both R&D intensity and the technology level, while a higher technology level is followed by smaller subsequent technological gains; when the lagged level is omitted, this negative convergence association loads onto the R&D coefficient. We term this mechanism "convergence masking" and hypothesize that adding a single convergence control recovers the within-country R&D-technology association. Data. Two country-year panels in long format covering the same 24 OECD countries: (1) main_panel_P1.csv, 2002-2019 (432 rows), where the technology variable is a principal-components composite of PCT international applications and triadic patent families; (2) wipo1b_panel_P1.csv, 2002-2022 (504 rows), where it is the WIPO 1b national-phase-entry patent count, used for robustness. All series derive from public sources: OECD Main Science and Technology Indicators (R&D intensity, gross R&D expenditure over GDP) and WIPO patent statistics. Variables are documented in README.md; lags were constructed within country. Notable findings. With country and year fixed effects and country-clustered standard errors (N = 408), the lagged R&D coefficient rises from 1.161 (p = 0.068) to 6.982 (p = 0.026) once the lagged technology level is added; the within R-squared rises from 0.001 to 0.104. The direction of this shift is robust to the alternative patent measure, to R&D lags of one to three years, and to leave-one-out country exclusion. Its magnitude is sample- and measurement-dependent (excluding Korea lowers the coefficient by about 73 percent) and is not part of the claim. Interpretation and use. Results should be read as a convergence-corrected within-country association, not a causal parameter, and as evidence on the direction of the bias rather than its size. Running P1_reproduce.R in R (4.0 or later) with the fixest and dplyr packages reproduces Table 1 and all robustness checks in the paper; expected outputs are annotated in the script and listed in README.md. Reported p-values come from fixest; other software may apply slightly different small-sample corrections to clustered standard errors, shifting p-values marginally while leaving coefficients unchanged. Economics Econometrics Innovation Sawng, Yeong-Wha Dataset doi:10.17632/m3jj8n89c8.1 oai:data.mendeley.com/m3jj8n89c8.1 Creative Commons Attribution 4.0 International http://creativecommons.org/licenses/by/4.0 https://data.mendeley.com/datasets/m3jj8n89c8 2026-06-12T02:45:20Z