NEWS

---

HOME 0.1.2

om_calibrate_luy() rewritten to reproduce Luy's calculation workbooks

The function was rebuilt from Luy's calculation workbooks (Multipliers_Italy_<year>_<SEX>_NEW_REF_YEARS.xls, which produce Tables A.5-A.20 of Luy 2009). Fed with Luy's inputs, the engine reproduces his reference periods and mean-age tables exactly and W(n) to within about 1 % up to respondent age group 45. Main changes:

• Inputs are four data frames in a fixed long format prepared by the user from any source: asfr (age, year, rate), cohort_qx (age, cohort, qx), pop_age (age, year, n), period_lx (year, age, lx or qx). The package does not read files.
• Parental survival is built, as in Luy, by averaging the cohorts' probabilities of dying with fixed weights (mixing = "luy"), not by mixing survival curves. mixing = "exact" gives the proper mixture.
• Unobserved ages are not filled in. Ages for which a parental cohort has no probability of dying (e.g. after the end of its series) are kept as NA and get no weight: the average in each year is taken over the cohorts observed in that year. Luy's workbooks filled these cells with q = 0 (1998) or q = 1 (2003); results differ from his published tables from respondent age group 50 onwards. If no parental cohort is observed in some year of exposure, W(n), a(n) and b(n) are NA with a warning; implausible W(n) are also set to NA with a warning.
• No re-levelling to the observed S(n): W(n) and t(n) come from the reconstructed survival only. The argument p_surv was removed.
• Exposure is n + 3 years; parental cohort = floor(T) - n - cohort_lag - x with cohort_lag = 4 (documented with reference to Luy 2009); reference period = T - K + mean death time measured at mid-year; period l(x) interpolated linearly between calendar years.
• Fertility: Schmertmann parameters read off the averaged ASFR (P, f(P), H), QS point densities at integer ages, mean age = sum (x+0.5) N f / sum N f, P/H stepped (+2/+1 up, -1/-2 down) and Brass-Gompertz fine-tuning. Paternal schedules are modelled on the female age scale and shifted by male_shift.
• New weighting ("Nf" = 2010 supplement / built-in tables, "f" = Luy 2009), cohort_lag, period_lx and smoothing arguments.
• a(n), b(n) from Brass-logit mortality scenarios S x 0.5-1.5.
• New output table mac: mean age at childbirth of all parents given the mean of surviving parents (Luy's Tables A.5-A.8).

New functions

• om_luy_data_needs() lists the years, ages and cohorts each input of om_calibrate_luy() must cover for a given survey date. Ages are capped at the open age interval of the life tables (max_age = 110).
• om_luy_validate_inputs() checks prepared inputs (columns, numeric whole ages/years, duplicated keys, ASFR per woman, q between 0 and 1) and, given a survey date, whether they cover the years, ages and cohorts needed.
• om_luy_intermediates() extracts every intermediate quantity of a calibration -- f'(x), N(x), weights, q[x,k], q-bar, L_k, share of observed weights and a per-mean-age summary -- laid out like Luy's workbooks.

Renamed and removed

• om_sensitivity_family() was renamed om_sensitivity_modelLT().
• The deprecated alias om_sensitivity() was removed; use om_sensitivity_Mn().

Other

• om_estimate_index(): a scalar mean_age_parent is recycled for all methods; length mismatches of p_surv / mean_age_parent are errors.
• tests/testthat.R now calls test_check("HOME"); previously no test ran during R CMD check.
• .qs_phi() returns 0 beyond the spline end point beta.