Currently, I am evaluating the model performance using Rsq. This may not be a useful metric. Potentially, if an ICS is always being predicted low, but the trend is right, that might be good enough. Here are some links the Rich suggested:
In addition to the above, when attempting to sell the model with users, what is the best way to describe its performance?
Do we somehow extrapolate the previous known outcome value and compare it with the predicted value? (this is useful only if this is something that our user would do perhaps?)
Do another method of calculating the predicted value using some other metric as a proxy for it?
Currently, I am evaluating the model performance using Rsq. This may not be a useful metric. Potentially, if an ICS is always being predicted low, but the trend is right, that might be good enough. Here are some links the Rich suggested:
General reviews
One backing up the R^2!
Directional accuracy
In addition to the above, when attempting to sell the model with users, what is the best way to describe its performance?