Class BossMissingDataStudy
MissingnessInjector with a chem-shaped rate profile.
Run this from IntelliJ; it streams a line per condition as it goes. With the defaults below it is a few minutes.
What this can and cannot show
MissingnessInjector is MCAR: the latent row propensity is independent of the data, so it clusters the
pattern without making deletion depend on any value. Under MCAR listwise deletion is consistent, so this
study compares methods on efficiency -- how much structure each recovers from the information that
survives -- and cannot show the bias differences that are the main statistical reason to prefer EM or multiple
imputation. Reading an EM win here as "EM corrects bias" would be wrong; it would be "EM uses more of the data".
The bias comparison needs the MAR term that is not yet in the injector.
The row propensity is the interesting dial precisely because it does not change the marginal rates: at a fixed level of missingness it moves cases between "many rows lightly hit" and "few rows heavily hit", and complete-case methods care enormously about which.
- Author:
- josephramsey
-
Constructor Summary
ConstructorsConstructorDescriptionDefault constructor for the BossMissingDataStudy class. -
Method Summary
-
Constructor Details
-
BossMissingDataStudy
public BossMissingDataStudy()Default constructor for the BossMissingDataStudy class.
-
-
Method Details
-
main
The entry point for the BossMissingDataStudy program, coordinating the execution of the missing data study under various conditions and configurations. This method performs two phases of data evaluation: one for continuous data and one for mixed data. Results are printed to the console to compare statistical efficiency across different methods used in the study.- Parameters:
args- Command-line arguments passed to the application. They are not used within this method but may be required for external execution.- Throws:
Exception- If any error occurs during the execution of the phases or data processing.
-