perudata 0.1.5

Peru's official INEI microdata (ENAHO, ENDES, EPEN, EEA, ENAHO Panel): download, load and validate against official statistics

perudata

Peru's official INEI microdata, one import away.

from perudata import enaho, validate

df = enaho.load(2024, "34")        # ENAHO sumaria: income, spending, poverty
validate.poverty(years=[2024])     # reproduces INEI's official 27.6% from raw data

Five national surveys, one consistent API. Every file is verified on download by actually opening it, and a built-in validation gate reproduces INEI's official poverty series for all 22 years, 2004–2025, at the precision INEI publishes it (one decimal) before you build anything on top. The unrounded residual never exceeds 0.05pp — that is the rounding of the published figure, not a disagreement, and the gate prints it rather than hiding it behind a rounded zero.

Survey	What it is	Coverage	Format
enaho	ENAHO — national household survey (income, poverty, employment, education, health, governance)	2004–2025, 29 modules	Stata
panel	ENAHO Panel — same households re-interviewed (true poverty dynamics, FE)	10 releases, 2007–2023	Stata
endes	ENDES — Peru's DHS (fertility, child health, anemia, domestic violence)	1996–2024	SPSS
epen	EPE/EPEN — permanent employment surveys (Lima monthly since 2001, national)	279 verified datasets	CSV
eea	EEA — annual economic survey of firms (sales, assets, employment, CIIU)	2001–2024, 604 datasets	CSV

Install

pip install perudata            # once published; for now:
pip install git+https://github.com/cesarchavezp29/perudata

Requires Python ≥ 3.9. Dependencies: pandas, pyreadstat.

Quickstart

from perudata import enaho, endes, epen, eea, panel, validate

# --- ENAHO (annual cross-section) ------------------------------------------
enaho.years()                        # [2004, ..., 2025]
enaho.modules()                      # DataFrame of the 29 modules
enaho.download([2023, 2024], ["01", "34", "85"])
df = enaho.load(2024, "34")          # auto-downloads if missing

# national poverty, the official INEI way (person = factor07 x mieperho):
w = df["factor07"] * df["mieperho"]
poverty = 100 * w[df["pobreza"].isin([1, 2])].sum() / w.sum()   # -> 27.6

# --- validation gate ---------------------------------------------------------
validate.poverty()                   # full 2004-2025 table vs official INEI

# --- ENAHO Panel (longitudinal) ----------------------------------------------
panel.releases()                              # [2011, 2015, ..., 2022, 2023] — 10 releases
df, meta = panel.load_long(2023, "sumaria")   # latest: 2019-2023, tidy long
meta["waves"]                                 # [2019, 2020, 2021, 2022, 2023]
# any release works the same, e.g. the earliest 5-year panel:
df, meta = panel.load_long(2011, "sumaria")   # 2007-2011

# --- ENDES (DHS) --------------------------------------------------------------
endes.download(2024, ["peso_talla_anemia"])
kids = endes.load(2024, "peso_talla_anemia")

# --- EPEN (employment) ---------------------------------------------------------
epen.search("lima")                  # find datasets in the verified catalog
df = epen.load(997)                  # download + read by catalog code

# --- EEA (firms) ----------------------------------------------------------------
eea.modules(2024)                    # sector modules of EEA 2024
df = eea.load(eea.modules(2024)["csv_code"].iloc[0])

Or from the command line:

perudata enaho --years 2024 --modules 01 34
perudata validate --years 2024
perudata epen --search "dpto 2024"

Where files go

Everything lands under ./peru_raw (override per call with out= or globally with the PERUDATA_DIR environment variable), named so you always know what a file is:

peru_raw/
  enaho/sumaria/enaho-2024-34.dta
  enaho_panel/2011_302/panel-2011-sumaria.dta
  endes/2024_968/1629_hogar/968-Modulo1629/*.sav
  epen/804_15_lima.../*.csv
  eea/987-Modulo1968/a2022_s04_fF2/*.csv
  */_manifest.csv          # what was downloaded, when, rows x cols

Downloads are idempotent: a file that exists and verified is skipped, so you can re-run a script after adding years and only the new ones are fetched.

Why this exists

INEI publishes superb microdata, but using it programmatically means knowing a pile of undocumented conventions, all of which this package encodes:

• Each survey-year is a numbered proyecto (1031-Modulo34.zip), and the codes are not chronological — code 279 is 2010 while 280–285 are 2004–2009. Every ENAHO code here was verified by reading the internal año variable of its file.
• ENAHO 2004–2011 ships Stata 7 (v110) files that pandas cannot read — load() falls back to pyreadstat transparently.
• Sumaria zips carry -12/-12g alternates that silently lack the poverty variables. The canonical file is selected for you.
• ENDES exists only in SPSS for the full 1996–2024 series, and its module numbers were renumbered in 2020 (64–74 → 1629–1641).
• The ENAHO Panel changed module numbering in 2018 (01/34 → 1474–1479) and old releases are WIDE files; panel.load_long() returns a tidy long panel either way.
• EPEN and EEA are CSV-only with no year→code formula at all — the package ships verified catalogs (279 and 604 datasets) discovered by probing the server and opening every file.

Validation

validate.poverty() recomputes national monetary poverty from the raw sumaria (person-weighted, factor07 × mieperho, pobreza ∈ {1,2}) and compares it to the official series from INEI's Evolución de la Pobreza Monetaria reports:

year  poverty_pct  official_poverty  pov_diff  poverty_exact  pov_diff_exact
2004         58.7              58.7       0.0        58.6990          -0.001
...
2024         27.6              27.6       0.0        27.5795          -0.021
2025         25.7              25.7       0.0        25.6667          -0.033

22/22 years match the published figure at INEI's own precision. poverty_exact is the unrounded value and pov_diff_exact the unrounded residual (largest: 0.049pp, in 2006) — reported on purpose, because rounding our own number before comparing would turn a real 0.03pp residual into a reported 0.0 and let the gate flatter itself.

Notes

• Data © INEI (Instituto Nacional de Estadística e Informática del Perú), published as open microdata. This package only automates download and reading — cite INEI and the survey when you publish.
• Value labels are not applied on load; you get the raw codes exactly as INEI documents them (the PDFs inside each zip are kept for EPEN/EEA).
• New ENAHO year out? enaho.discover(2026) scans the server for its code.

License

MIT — see LICENSE. Built by Carlos Chávez Padilla.