Using residuals to identify model families? #223
-
|
Residuals can be used to choose between model families (gamma, etc), but are they also useful for full model comparison, or is it better to lean more on cross validation and the other techniques you mentioned? |
Beta Was this translation helpful? Give feedback.
Answered by
ericward-noaa
May 12, 2023
Replies: 1 comment
-
|
Partly depends on goal: prediction vs. inference. If purely prediction, cross-validation is always best. If inference is also goal, then want model to be consistent probabilistically with the data. I.e., you want the residuals to look reasonable and to be only comparing models that are consistent with the distributional assumptions. |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
ericward-noaa
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Partly depends on goal: prediction vs. inference. If purely prediction, cross-validation is always best. If inference is also goal, then want model to be consistent probabilistically with the data. I.e., you want the residuals to look reasonable and to be only comparing models that are consistent with the distributional assumptions.