The Long Shadow of Missing Women on Marriage and Fertility

Description

Replication package for "The Long Shadow of Missing Women on Marriage and Fertility" (Sunha Myong), Economic Modelling. HYPOTHESIS. Sex-selective abortion in South Korea in the 1980s and early 1990s raised the sex ratio at birth from about 1.05 to about 1.15. The hypothesis is that this reduces marriage and, through marriage, fertility with a lag of roughly three decades, as the affected cohorts reach prime marriage age. DATA. Three Statistics Korea microdata products from the Microdata Integrated Service (MDIS, mdis.mods.go.kr): Census 2% samples 2000-2020, marriage registers (B-type) 2001, 2006, 2011, 2016, 2021, and the 2020 birth register. All three are public-use files, downloadable directly by any registered MDIS account without application, approval or fee. Also WDI, UN WPP 2024 and KOSIS tables. Individuals fall into 24 types by education, age band and capital region; the estimation uses, per vintage, the 24x24 matrix of marriages by (husband, wife) type, never-married counts and the implied Choo-Siow surplus. These are included in full. SYNTHETIC DATA. The MDIS licence does not permit redistribution, so the three files are not included; DATA_AVAILABILITY.md documents the download procedure and variable selection. Synthetic replacements are provided instead: artificial records generated to reproduce a dataset's structure while containing none of its records. Each variable was regenerated from a one-way frequency table, so a replacement matches its original in file name, variable names, observation count, missing counts and every distribution. Variables are drawn independently, so each marginal distribution is exact and the joint structure is not. In the 2021 register 78% of college-educated husbands have a college-educated wife; in the synthetic file the figure is 58%, the value implied by independence. That sorting is the object of the estimation; a synthetic run establishes only that the code executes. FINDINGS. Restoring the women missing from the 1980-1995 cohorts, relative to a natural sex ratio of 1.050, raises the 2021 female marriage-market stock by 5.53% and lowers the market sex ratio from 1.243 to 1.178. Re-solving the equilibrium with the surplus held fixed yields 5,008 additional marriages, 3.23% of the 154,890 observed in 2021, and about 8,500 additional births, 3.3% of the 2021 total. The effect is lagged and increasing: 0.86% in 2006, 1.35% in 2011, 2.28% in 2016, 3.23% in 2021. College-educated women aged 25-34 are 42.4% of the restored stock and 74.5% of new marriages. USE. The counterfactual is partial equilibrium: stocks vary while the surplus is held at baseline. Stage 1 (Stata) constructs the 24-type matrices from the microdata; stage 2 (MATLAB) estimates the surplus, solves the counterfactual and produces every table and figure. Because the matrices are included, stage 2 reproduces every published result without microdata. README.md documents the procedure, software versions and exhibit-to-script mapping.

Steps to reproduce

SOFTWARE. StataNow/SE 19.5 (each do-file declares version 17.0, so Stata 17 or later runs it unchanged; any edition), MATLAB R2025b using base MATLAB only (no toolboxes), and Python 3.10 with numpy and pandas, needed only for the synthetic data. No user-written Stata or MATLAB packages are required. 1. OBTAIN THE MICRODATA. Two Statistics Korea products are needed, both public-use files from the Microdata Integrated Service (MDIS, mdis.mods.go.kr): the Population and Housing Census 2% samples for 2000, 2005, 2010, 2015 and 2020, and the Vital Statistics marriage registers (B-type) for 2001, 2006, 2011, 2016 and 2021. Register a free account, select the variables listed in DATA_AVAILABILITY.md section 1.2, and download. The MDIS licence bars redistribution, which is why the files are not in this package. Place them in data/raw/census/ and data/raw/marriage_vital/ under the names given in each folder's READ_ME_FIRST.md. 2. SET ONE PATH. In code/stata/config.do, edit the single line defining REPROOT to the absolute path of the unpacked package. Every other path derives from it; no other file needs editing. 3. RUN THE STATA STAGE. Execute code/stata/master.do. It imports each marriage register, harmonises codes across vintages, cleans the census extracts, assigns every individual to one of 24 types formed by education (four-year college or more versus less), six age bands (15-24, 25-29, 30-34, 35-39, 40-44, 45-49) and capital region (Seoul, Gyeonggi, Incheon versus elsewhere), and exports per vintage the 24x24 matrix of marriages by husband and wife type, the never-married stocks and the population by type. Import and cleaning are the slow steps; set run_import and run_clean to 0 to skip them on a re-run. 4. RUN THE MATLAB STAGE. From code/matlab, run master.m. It recovers the Choo-Siow (2006) marital surplus as Phi(i,j) = log mu(i,j) - 0.5 log mu(i,0) - 0.5 log mu(0,j), then restores the women missing from the 1980-1995 birth cohorts by scaling each age band's female stock by its cohort sex ratio at birth divided by the natural benchmark of 1.050, holding Phi and the male stock fixed, and re-solves the matching equilibrium by iterative proportional fitting to a tolerance of 1e-10 with a cap of 20,000 iterations. 5. COLLECT RESULTS. Tables are written to output/tables/. README.md section 6 maps every table and figure in the paper to the script that produces it. REPRODUCING WITHOUT THE MICRODATA. The 24-type matrices ship with the package, so step 4 alone reproduces every published table and figure; steps 1 and 3 are needed only to rebuild those matrices from source. To confirm the Stata code executes, set usesynthetic 1 in config.do to run steps 3 and 4 against the synthetic replacements in data/synthetic/. That run demonstrates execution only; it does not reproduce the published numbers.

Institutions

• Kyung Hee University

  Seoul, Seoul

Categories

Funders

• Kyung Hee University

  Grant ID: 20252809

Licence