postpi

postpi corrects biases in post-prediction statistical inference by modeling the relationship between observed and predicted outcomes to improve variance estimation for downstream analyses.

Key Features:

• Bias Correction: Identifies a low-dimensional representation of the relationship between observed and predicted outcomes to correct bias in statistical inference.
• Variance Estimation Improvement: Models the relationship between observed and predicted outcomes to enhance variance estimation for more reliable inferential results.
• Integration with Machine Learning Frameworks: Operates with standard training, testing, and validation set workflows commonly used in machine learning analyses.

Scientific Applications:

• Genomics: Improves inference in modeling predicted phenotypes using re-purposed gene expression data.
• Public Health: Enhances accuracy in predicting causes of death from verbal autopsy data.

Methodology:

A prediction model is trained on a training set; the relationship between observed and predicted outcomes is estimated on a testing set; the estimated relationship is applied to correct inference on a validation set, improving variance estimates.