Replication material for "The Limits of Predicting Individual-Level Longevity"

Folder structure

1_raw_data

• The raw harmonized Health and Retirement Study (HRS) data used in this study are publicly available and can be downloaded here

2_long_data

• make_long_data.R - Script to convert the Raw HRS data from wide to long format, saved as hrs_long.rds
• HRS_cleaning_all.R - Script to clean the data, select the variables on interest, and produce hrs_long_cleaned.rds
• HRS_variables_list.R - List of all variables, levels, data types included in the analyses

3_cleaned_variables

• hrs_long_cleaned.rds - Output of HRS_cleaning_all.R

4_modelling data

• make_modelling_data.R - Script to produce the training data train.rds and the testing data test.rds
• train.rds - Training dataset for training all models (60% of full sample). Output of make_modelling_data.R
• test.rds - Test dataset for evaluating all models (40% of full sample). Output of make_modelling_data.R

5_predictions

• models.ipynb -

• models.R -

• curves - Folder that contains the predicted survival curves

• model_objects - Folder that contains the predicted model objects

• oversampled - Folder that contains the predicted survival curves and model objects for the oversampled data

6_evaluation

• evaluate_predictions.ipynb - Script to compute the prediction accuracy metrics
• evaluation_visualization.R - Script to visualize the prediction accuracy metrics
• variable_importance.py - Script to calculate variable importance for the nonparametric models
• all, educl, gender, race - Folders that contains the prediction accuracy metrics for the main test dataset, and the oversampled dataset for education level, gender, and race and ethnicity

7_Expectations

• expec_final.R - Script to compute subjective survival probability

8_outputs

• Figures included in the paper