search

mankoff / data-practices

Some suggested data practices
Creative Commons Zero v1.0 Universal
1 stars 0 forks source link

Plotting "Observed vs. Predicted" or "Predicted vs. Observed" #2

Open mankoff opened 2 years ago

mankoff commented 2 years ago

PO (Predicted x, Observed y) is incorrect, and you should plot OP (Observed y, Predicted x), according to https://doi.org/10.1016/j.ecolmodel.2008.05.006

niccolodematteis commented 2 years ago

The results of figure 2 derive directly from the definition of predicted (P) and observed (O) data. In particular, P=O+err, while P has no error. I am not familiar with ecological modelling, but I would adopt the opposite definition, i.e., O=P+err. That is, the model is affected by some form of uncertainty, while the observations are "correct". Using this convention to plot the data implies that the correct way is PO (predicted y, observed x). In the end, I think that the choice between PO and OP depends on what are you presenting and what information you intend to communicate

mankoff commented 2 years ago

Thanks for commenting here - I agree, it is often reasonable to assume model has error and observations are correct. Depends on the situation.

In the case of "validating my new product against existing 'standards'", maybe assuming that the new product has errors is reasonable (although, of course, historical standards also have errors).

mankoff commented 2 years ago

Although, methods exist to deal with errors and uncertainty in both the dependent and independent variable. Seems like those are the better methods to use.