Old-Age Mortality Smoothing

Project Overview

Purpose: This project evaluates various statisical methods for smoothing old-age mortality rates in the Human Mortality Database.

Models Tested: Current methods explored include classic parametric models (Beard, Kannisto, Gompertz, Makeham, Log-quadratic), mixture models, and p-splines. Methods in development include Bayesian hierarchical models and machine learning apporaches.

Evaluation Metrics: AIC is metric used for assessment currently. Planning to expand to cross-validation metrics like MSE/accuracy.

Repository Strucutre

Data: Raw data are mostly downloaded input data files from mortality.org. Processed data include raw death data and exposure data calculated using the extinct cohort method. More to come here, working on organizing data files.

Code: Code files include data preparation, EDA, and function scripts. Also includes the current draft of the manuscript.

Data Preparation

code/create_cohort_validation_data.qmd

code/prep_france.qmd

code/prep_scandi.qmd

Exploratory Data Analysis

code/eda.qmd

Model fitting functions

code/model_fitting_functions_cohort.R

code/local_flex_models.R

Manuscript

code/manuscript.qmd

Figures: Charts and tables generated during analysis.