NEWS

---

incomeLato 0.1.1

All console messages, warnings and function help are now in English. Exclusion reason codes renamed (see README). No change to predictions.

• motivo_exclusao codes: venezuela_quarentena_arquitetural → venezuela_architectural_quarantine, pais_fora_dos_15_ativos → country_not_in_15, fora_da_janela_temporal_valida_do_pais → outside_country_valid_window, preditor_ausente_ou_invalido → missing_or_invalid_predictor. Code that filters on the old values must be updated.
• Result attributes uso_recomendado / uso_proibido renamed recommended_use / prohibited_use, with English values.
• The usage notice (warning, README, vignette) now separates what is allowed from what is not: comparing groups across waves is allowed (e.g. party support among the bottom 50% in each wave); reading mean(prob_bottom50) as the size of the bottom half, or its change as a trend, is not.
• Column names are unchanged, including veredito_deriva_temporal and flag_margin, so existing code keeps working. Moving column names to English is left for a future version, with a deprecation warning.
• Validation, description corrected (numbers unchanged). The README and the vignette now separate the validation against LAPOP observed income (the household income bracket, compared respondent by respondent) from the validation in use (party support computed with observed vs. predicted income weights). Two descriptions in 0.1.0 were wrong: the 339 v1.4 comparisons are country-wave-party, not country-wave; and the 97% top-party agreement under v1.5 is over 66 country-waves, not over the 357 country-wave-party comparisons.
• Validation section rewritten around the packaged recipe (script 265 of the research project): the v1.5 model with the v1.4 recalibrators, Mexico uncalibrated, against LAPOP observed income, 66 country-waves, 2008–2018, in-sample for the recalibrators. The numbers cited in 0.1.0 (0.21–0.42, 0.97–0.99, 0.670 pp, 97%) came from the v1.4 and v1.5 models without recalibration; the vignette keeps them in a comparison table, labeled as such.
• README and vignette example rewritten: it now goes through income_lato() on a file and shows support for democracy (simulated) in the three income groups by country.
• The README no longer says that income_lato() returns "your data": the other columns of the input file are not carried over, and respondent_row_id (the row's position in its file) is the key to bring them in.
• Checked against v0.1.0 on the real 2018 wave: 16,901 of 16,901 predictions identical (maximum difference 0).

incomeLato 0.1.0

First public release.

Archived on Zenodo: DOI 10.5281/zenodo.23045517 (this version); 10.5281/zenodo.23045516 (all versions).

• Applies the frozen income-position model v1.5 (15 countries: Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Ecuador, El Salvador, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay) to Latinobarómetro waves 1995–2024 supplied by the user, in .dta or .sav.
• income_lato() accepts a single wave file or a folder of waves, and ends with a message summarizing waves detected, waves excluded by default, respondents scored, and countries with no prediction and why.
• Recalibration: v1.4 recalibrators for 14 countries; Mexico uncalibrated. This is the combination behind the published article's predictions.
• The package ships no survey data and nothing computed from survey microdata: model parameters, variable-name crosswalks, excluded country-waves, and per-country categorical verdicts only.
• The 2024 wave is now readable (codes are read as numbers instead of labels); it stays excluded by default because its education codes were inferred, not confirmed by labels.

Known limitations

• The per-class validation (bottom 50 / middle 40 / top 10) cited in the README is still the one run under model v1.4; redoing it under v1.5 is planned for v0.1.1. The aggregate party-support validation is already under v1.5.

Validation

End-to-end check of the installed package against the article's predictions (data-raw/golden_file_validation_v1_5.R, 2026-09-29): 369,999 of 369,999 respondents matched (waves 1995–2024, 15 countries); maximum absolute difference 0 and correlation 1.0 for all three probabilities in every country; predictors and weights identical row by row. One deliberate difference in a flag: flag_piso_confianca for Mexico is FALSE in the package and NA in the article's file.