Model evaluation, validation and verification Reading scientific articles that have a statistical model, I sometimes find these three terms:

*

*Model evaluation

*Model validation

*Model verification

From a statistical point of view, are all these terminologies existing/correct? If yes, did they mean the same thing (can they be used interchangeably)?
I am referring particularly to the meaning of the term and not the author's intended meaning.
I always get puzzled when I found these kinds of things (sometimes I have the feeling that there is a sort of "poetic license" when talking about statistic).
 A: In my view, the term "model verification" is misleading and should not normally be used. Models are by their very nature not "really true", and the term seems to suggest that one could somehow make sure that they are, which is impossible.
"Model evaluation" and "model validation" can be used more or less synonymously, however I'd interpret "model evaluation" as more exclusively quantitative, whereas "model validation" may involve use of visual tools and also background information. "Model validation" seems to address a rather binary question whether a model is "valid" or not to confirm or rule out its further use, although in fact of course "valid/invalid" is not strictly binary and there is a gradual transmission between them. "Model evaluation" doesn't imply a decision on whether to further use the model or not, and can be done to compare different models all of which (or none of which) can be seen as "valid".
However, these aren't sharp delimitations of the term, and I wouldn't call "wrong" the use of "model evaluation" for what I can "model validation" and the other way round. In order to clarify things authors should explicitly state what they mean by these, particularly if they want to exclude certain uses.
