Skip to contents

Changelog

Source: NEWS.md

OECDscaleup 0.1.0

Initial release: a behaviour-preserving translation of the OECD Scale Up project’s “Second Package” Stata pipeline (MASTER_OECD_p2.do and its 5 worker .do files, version 1.0, scaleup@oecd.org) into R.

What this release contains

  • config_scaleup() — replaces the Stata global macros set in MASTER_OECD_p2.do.
  • build_accounts_indicators() — translates n0_1_accounts_indicators_v2.do.
  • compute_characteristics() — translates n1_3_characteristics.do.
  • compute_persistence() — translates n1_3_persistence.do (transitions + evolutions).
  • compute_contribution() — translates n1_3_contribution.do.
  • compute_predictors() — translates n1_3_predictors.do.
  • run_scaleup_pipeline() — file-I/O orchestrator mirroring MASTER_OECD_p2.do’s role; every function above it is a pure, independently-testable data transformation.
  • Bundled reference data tl3_typology and exchange_rates, carried over as-is from the OECD-supplied .dta files.
  • A documented country_patches lookup table replacing the 3 hardcoded country-list branches in the original n0_1_accounts_indicators_v2.do.

Translation notes

  • Ported from Pseudocode/*.md (a structure-preserving intermediate representation of the original Stata source), not directly from the .do files — see that directory for the line-by-line correspondence.
  • Several Stata/R semantic gaps were identified and deliberately corrected for, each with a dedicated regression test: Stata’s non-propagating missing-value comparisons (stata_num()), replace ... if semantics under an unresolved condition (stata_replace()), Stata’s round-half-away- from-zero convention (round_half_up()), and Stata’s percentile algorithm (percentile_stata(), confirmed equivalent to R’s quantile(type = 2)).
  • The original has no automated test suite; the one existing behavioural spec (the confidentiality worked example in the source PDF, §4) informed this package’s dominance-share tests. Everything else is covered by new unit and integration tests built from synthetic fixtures, since the real firm microdata this pipeline processes is confidential and never ships with the package (see Analysis/stata.md and variable_dictionary_v1.xlsx for the input schema fixtures are built against).
  • No .dlg dialog files exist in the source package, so no Shiny GUI was built (there was nothing to recreate).
  • Known residual approximations, not exact-match verified against a live Stata run: statar::xtile()’s tie-breaking at quartile boundaries, and fixest’s robust-standard-error small-sample adjustment. Both are flagged in the relevant function’s documentation.