Demographic and life-table forecast dataset for Uzbekistan, 2020–2050: a hybrid Lee–Carter–Gompertz mortality-modelling approach
Description
This dataset provides abridged life tables and mortality indicators for Uzbekistan at the national level, covering historical reference years (2000, 2005, 2010, 2015, 2019) and annual forecasts for 2020–2050. Seven life-table functions are reported — central death rate (nMx), probability of dying (nqx), survivors of a radix of 100,000 (lx), deaths (ndx), person-years lived (nLx), total person-years remaining (Tx) and life expectancy (ex) — across 19 abridged age groups (<1, 1–4, 5–9, …, 80–84, 85+). Historical indicators are given for both sexes, males and females; forecasts are given for males and females. Historical rates are drawn from the WHO Global Health Observatory. The 2008 all-population empirical life table (single year of age) is from the State Committee of the Republic of Uzbekistan on Statistics (www.stat.uz). Forecasts were produced with a hybrid Lee–Carter–Gompertz model in which the log central death rate is expressed as ln m_x(t) = (ln A + B·x) + a_x + b_x·k_t + ε, with the mortality index k_t extrapolated by an ARIMA(1,1,1) process. Data are provided in tidy (long) CSV format keyed by year × sex × age group × indicator, together with a multi-sheet Excel workbook, two single-year reference life tables for 2008 (empirical and Gompertz), a model-comparison table (R²/RMSE by model and sex), a data dictionary, and a reproducible Python build script. Each value carries a provenance flag: "published" values are transcribed directly from the source, while "derived" forecast values (ndx, nLx, Tx) are reconstructed from published lx using the standard abridged life-table algorithm. All life-table identities were validated programmatically (0 ≤ nqx ≤ 1; monotone lx; ndx = lx(x) − lx(x+n); ex = Tx/lx); a validation report is included.
Files
Steps to reproduce
1. Inputs. Abridged life-table indicators for Uzbekistan (both sexes, male, female) for 2000, 2005, 2010, 2015 and 2019 were obtained from the WHO Global Health Observatory. The 2008 all-population empirical life table (single year of age) was obtained from the State Committee of the Republic of Uzbekistan on Statistics (www.stat.uz). 2. Mortality model. Age-specific log central death rates were modelled with a hybrid Lee–Carter–Gompertz specification, ln m_x(t) = (ln A + B·x) + a_x + b_x·k_t + ε, where (ln A + B·x) is the Gompertz baseline, a_x the average age-specific mortality level, b_x the age-specific response to change, and k_t the time-varying mortality index. Parameters a_x and b_x were estimated by singular value decomposition of the centred log-rate matrix, under the identification constraints Σ b_x = 1 and Σ k_t = 0. Projection. The mortality index k_t was extrapolated to 2050 with an ARIMA(1,1,1) model (φ ≈ 0.7, θ ≈ 0.3, σ ≈ 0.005), and the projected log rates were back-transformed to central death rates nMx. 3. Life tables. Abridged life tables were constructed from nMx using separation factors a0 = 0.1 (age <1), a = 1.6 (ages 1–4) and a = 2.5 (five-year groups), with nqx = n·nMx / (1 + (n − a_x)·nMx). The open-ended 85+ interval was closed using the model life expectancy e(85+). 4. Packaging and QA. The included Python script (code/build_dataset.py) reshapes all tables into tidy CSVs, reconstructs forecast ndx, nLx and Tx from published lx, builds the Excel workbook, and runs a life-table validation suite (see validation_report.txt). Goodness of fit is summarised in reference_model_comparison.csv.
Institutions
- National University of UzbekistanTashkent, Tashkent