Mukhtar-Data/somaliland-early-marriage-shds-2020

Early marriage among ever-married women in Somaliland using the 2020 Somali Health and Demographic Survey. # Early Marriage Among Ever-Married Women in Somaliland, SHDS 2020 ## Research title **Early Marriage Among Ever-Married Women in Somaliland using the 2020 Somali Health and Demographic Survey.** ## Background This study uses the 2020 Somali Health and Demographic Survey (SHDS) to examine early marriage among ever-married women living in Somaliland. It combines multilevel regression with cluster-aware machine-learning prediction and model interpretation. ## Research objective To examine characteristics associated with early marriage and to evaluate how well the prespecified study variables predict early marriage. ## Study population The analytical population comprises ever-married women aged 20-49 years residing in Somaliland. The final complete-case analytical sample contains **3,869 women nested within 140 survey clusters**. ## Data source Secondary data from the **2020 Somali Health and Demographic Survey (SHDS 2020)**. The manuscript states that access to the microdata is provided by the Somali National Bureau of Statistics microdata repository after registration and approval. ## Outcome definition Early marriage is first marriage before age 18 years. It is coded as a binary outcome: `1` for marriage before age 18 and `0` for marriage at age 18 or older. ## Predictors The study includes demographic, socioeconomic, geographic, reproductive, household, media-access, digital-access, and women's decision-making characteristics. The 20 prespecified predictors are listed in docs/data_dictionary.md. ## Statistical methods The Stata workflow produces descriptive statistics, bivariate logistic regression, a multivariable logistic model, and random-intercept multilevel logistic models clustered by survey cluster. Candidate variables used a documented bivariate screening criterion of *p* < 0.25. Multicollinearity was assessed with variance inflation factors; model fit, ICC, and stability diagnostics were also examined. ## Machine-learning methods The lock …