Timeline for Can I retain the ordinal nature of a predictor while answering a question about it that is inherently binary?
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 27 at 19:29 | comment | added | Frank Harrell | There is almost no limit for how noisy a variable can be before dichotomization doesn't make it worse. Translating a quantitative error to a qualitative error is not a good result. | |
| Aug 27 at 16:42 | comment | added | jginestet | @FrankHarrell, my suggestion to dichotomize was under the assumption that the predictor came from a Likert scale. Per the OP, it is not... So there... I also never stated that "because you will dichotomize in the end, you can do so at the beginning". Just that Likert scales are such poor instruments that you do not lose any information by dichotomizing from the beginning. Since here the poredictor is not a Likert... | |
| Aug 27 at 11:48 | comment | added | Frank Harrell | The link I provided was primarily about continuous variables but almost all of it also applies to ordinal variables. And thinking that the need for a yes/no answer translates into up-front dichotomization is a major mistake though a very common one. Never categorize on the front end; do so only on the back end if at all. | |
| Aug 27 at 10:40 | comment | added | Nick Cox | I won't downvote this because you're mentioning that the procedure is bad practice, yet I have to join a chorus that throwing away some information in the data is always to be avoided if possible. | |
| Aug 27 at 10:24 | comment | added | mkt | But FWIW (i) this is not a Likert scale, and (ii) I tried your suggestion. I fitted two models with the same predictor, one on the ordinal scale and the other discretized. By LOOCV, the ordinal model performed substantially better. | |
| Aug 27 at 10:24 | comment | added | mkt | Thank you for this. In general, I think it is helpful to respect the properties of the data when fitting the model. A binary decision does not necessarily require the data itself to be discretised. But how to get from the ordinal model to the binary decision is not obvious, which is where I seek input. | |
| Aug 26 at 23:35 | comment | added | jginestet | @FrankHarrell, the link you provided is about categorizing continuous variables. My answer agrees that this is poor practice. So why the disagreement? I however note that dichotomous diagnostic tests (Covid, TB, pregnancy, etc.) all do exactly that; transform a continuous outcome (typically a level of analyte(s) of interest), and transform this into a yes/no answer; yes you are pregnant, or no you are not, based on the (in that case) level of hCG in urine (which is an analog, continuous quantity). So I agree that it is generally bad practice, but in some cases, it can serve a purpose. | |
| Aug 26 at 22:06 | comment | added | Frank Harrell | This is very problematic and does not represent best statistical practice. There is no threshold that will be reproducible, and a host of problems as detailed here | |
| Aug 26 at 21:47 | history | edited | jginestet | CC BY-SA 4.0 |
edited typos in body
|
| Aug 26 at 20:02 | history | edited | jginestet | CC BY-SA 4.0 |
Added comment re. NPS
|
| Aug 26 at 19:51 | history | edited | jginestet | CC BY-SA 4.0 |
Reworded a paragraph
|
| Aug 26 at 19:45 | history | answered | jginestet | CC BY-SA 4.0 |