The authors state the chose median and mean absolute error in frequency to evaluate model performance, but don’t give proper justification for why this is the correct choice of metrics
TODO:
Interest here is in probabilistic forecasts.
Say that generally we’re interested in proportions and have large enough sample counts that it’s mostly fine and there is low uncertainty due to sampling (?)
But point to new supplementary figures for coverage for comparison
The authors state the chose median and mean absolute error in frequency to evaluate model performance, but don’t give proper justification for why this is the correct choice of metricsTODO: