What is the difference between external and internal cross validation? What is the difference between external and internal cross validation? From here I found:

Internal statistical cross-validation assesses the expected
performance of a prediction method in cases (subject, units, regions,
etc.) drawn from a similar population as the original training data
sample. Internal validation is distinct from external validation, as
the latter potentially allows for the existence of differences between
the populations (training data, used to develop or train the
technique, and testing data, used to independently quantify the
performance of the technique).

But I still don't quite understand the difference.
 A: The exact meaning of internal vs. external validation depends on context.

In study design e.g. in psychology,

*

*internal validation refers to finding out how well the available data set supports the answers to the studied questions. I.e., validity within what can be said from the particular data.


*external validation refers to the validity of extending the answers/conclusions from the study to other data acquired outside the data set available in the study.
In the context of model validation, that would mean that e.g. for a model that predicts some disease outcome, external validation tests how well the model predicts data acquired at hospitals in other countries, for patient populations with different demographics, ...

In analytical and clinical chemistry,

*

*internal validation are all the validation steps a lab does for a method they establish.
Methods here are procedures (both wet lab + data analysis/prediction) that yield e.g. analyte concentration or some diagnostic score from a sample.


*external validation means that the test samples are physically prepared by an external organization (e.g. a ring trial organizer). These samples are then sent to the lab who wants or needs an external validation for their method, they treat this sample like any other "production case/sample" and send their prediction/analysis result back to the external organization. The external organization then compares the lab's prediction with their reference value and certifies
You may say that external validation in this context is like a double blinded validation study whereas internal validation is at best single blinded (if at all blinded).


*

*Cross validation is internal in both senses, since the test cases are randomly selected from the same data set on which training is performed.


*A validation study testing a ready-to-use method with new cases performed by the lab itself may be addressing external validity in the study design sense, but would still be considered internal in the analytical chemistry sense.


*External validation in the analytical chemistry sense is also about a particular aspect of external validity in the study design sense.
