Closed finncatling closed 3 years ago
Also note that all models are tested only on complete cases
Proposal for test cases:
Evaluate model performance using all test cases including incomplete cases. Comparison with NELA model would be on complete cases only.
Difference in our model performance between the two would perhaps give a clue if there are systemic differences between complete and incomplete.
Decision with @JMathiszig-Lee to compare current-NELA-model complete and incomplete cases by comparing mortality in these two groups.
Now this comparison happens in the demographic table
We use the just the complete cases for training the current model. We use complete cases plus the incomplete cases for training the novel model.