Timeline for Does it make sense to compare different imputation techniques?
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 2, 2023 at 9:42 | vote | accept | olke | ||
| Jun 29, 2022 at 17:17 | comment | added | olke | Ah now I see, you meant choosing the imputation method that yields the values most 'desirable' for the study outcome. Makes perfect sense. Thanks for the answers you both, very helpful! | |
| Jun 29, 2022 at 16:16 | history | edited | David L Thiessen | CC BY-SA 4.0 |
added 7 characters in body
|
| Jun 29, 2022 at 15:48 | comment | added | David L Thiessen | @jbowman 's second definition is what I was thinking of, I've edited to clarify. Trying different imputation methods to find the most accurate may lead to overfitting the data, but it's probably balanced out by the more accurate imputation values. Hard to say. Train-Test splitting or Cross Validation can be used as well, but get computationally prohibitive. | |
| Jun 29, 2022 at 15:35 | history | edited | David L Thiessen | CC BY-SA 4.0 |
added 22 characters in body
|
| Jun 29, 2022 at 13:52 | comment | added | jbowman | That's the "good" definition of "most favourable"; there's another definition - "gives me the best-looking study results" - that is akin to p-hacking. | |
| Jun 29, 2022 at 11:07 | comment | added | olke | Thanks for such an elaborate answer! As a very fresh PhD student, I had no idea it may be considered shady. Could you please elaborate on why it's the case? I was under the impression that finding the most favourable imputation method (i.e. the one that yields the most accurate imputed values) is desirable (as opposed to p-hacking). | |
| Jun 28, 2022 at 16:58 | history | answered | David L Thiessen | CC BY-SA 4.0 |