1

From simulation studies, I have repeated (1000) measures of the root-mean-squared-error (deviation) between a sample of observed values and the predicted value. The obtained RMSE values are, naturally enough, all slightly different. The samples contributing to the 1000 measures (estimates) are all of the same size and relate to the same predicted value.

How can I use those obtained values to obtain an estimate of the population RMSE? Does it make sense to use the mean and standard deviation of the obtained RMSE estimates to describe what is, in effect, a series of 1000 estimates of the RMSE, each based on identical model parameters? Perhaps an alternative to that question is, "Is it patently a stupid thing to do?"

CrimsonDark
  • 239

1 Answers1

1

If they're all estimates of the same population RMSE, I'd be inclined to average mean-square errors (take means of means) and then take the square root at the end.

If they're for different sized simulations then weighted averages of MSEs would make sense.

Glen_b
  • 282,281