We have two treatments A and B. For A, we have 36 paired samples (measured vs simulated) and for B, we have 27 paired samples. We want to check the performance of APSIM for both the treatments, so we worked out coefficient of determination (r2), slope (b), root mean squared error (RMSE), relative RMSE and Wilmot's index of agreement (D). Do you think, we are doing biased analysis because we have heterogeneous number of samples for two treatments? Please reply or any other suggestion for improvement.
Cheers
Raman
You should normalise the RMSE by the mean value of the measurements = RMSE/mean
Hi Leonard
I am using relative RMSE or normalized objective function.
Relative RMSE is popular in such cases. If you are not happy with that. You may want to try other options. One of that could be testing your model's efficiency separately for each treatment. This way you can avoid getting biased towards more numbers of paired samples.
For more information on model efficiency: Nash JE, Sutcliffe IV. 1970. River flow forecasting through conceptual models. Part I – A discussion of principles. J. Hydrol. 10, 282-290.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis