Timeline for How to choose between ordered logit and ordered probit regression?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 15 at 16:22 | history | bounty ended | kmm | ||
| Aug 15 at 11:22 | comment | added | Graham Wright | @DrJerryTAO, thanks so much for that clarification. I had no idea about the "FrankHarrell" conception, and had only ever heard of the "dimitriy" conception, either marginal effects at the mean (where you hold other covariates at the mean and increase they key IV by 1) or average marginal effects (where you hold other covariates at the specific value of each observation, and average the effect across the entire dataset). No idea that there was such confusion about this term. | |
| Aug 12 at 21:20 | comment | added | DrJerryTAO | The disparity in opinions comes from different definitions of marginal effects. What FrankHarrell calls "marginal effects" are odds ratios after removing most predictors and estimating another model with only one predictor for marginal relationship. What @dimitriy calls "marginal effects" are differences in probabilities when manipulating the treatment variable while fixing covariate settings at certain levels (usually at representative combinations, mean of each covariate, or over the observed distribution in the sample). The Health Services Research must have referred to the latter. | |
| Aug 12 at 11:47 | comment | added | Frank Harrell | Marginal effects are not what analysts think they are: fharrell.com/post/marg and odds ratios are likely to transport outside the study to the extent that either interactions are properly handled or the distribution of interacting factors is the same. Covariate-specific difference in risk are needed: fharrell.com/post/rdist | |
| Aug 11 at 23:54 | comment | added | Graham Wright | IMO, ORs are one of the most commonly misinterpreted concepts statistics, with even published papers in reputable journals frequently misinterpreting an OR of (say) 1.25 as meaning "25% more likely." Whatever utility they have to people who do understand them is (to me) far outweighed by their tendency to confuse those who do not. So to me, the fact that logit allows odds ratios doesn't count as a point in its favor. Logit and probit both provide obviously uninterpretable coefficients, and can both produce "wrong but useful" marginal effects. So I agree that both are equally good. | |
| Aug 11 at 21:30 | comment | added | DrJerryTAO | @FrankHarrell do odds ratios transport outside of they study? I do not think so. What do "marginal effects" refer to in each of "Marginal effects are highly problematic", "estimated differences in probabilities for selected covariate settings", and "marginal estimates using arbitrary covariate distributions?" Researchers usually give adjusted predictions, marginal effects at the mean, or average marginal effects. Why are marginal effects are problematic? Are odds ratios better in these aspects? | |
| Aug 10 at 15:16 | comment | added | Frank Harrell | That article and the HSR editorial policy are way, way off the mark IMHO. I’ve written multiple blog articles about this at fharrell.com . Marginal effects are highly problematic and don’t transport outside of the study. And if you think ORs are hard to interpret, just try to interpret a regression coefficient in a probit model. Note that it is a great idea to provide estimated differences in probabilities for selected covariate settings. That’s far different from marginal estimates using arbitrary covariate distributions though. | |
| Aug 10 at 12:33 | history | edited | dimitriy | CC BY-SA 4.0 |
deleted 43 characters in body
|
| Aug 9 at 21:31 | comment | added | dimitriy | @FrankHarrell I find it much harder to think in terms of odds and prefer marginal effects on probabilities, which are the norm in my field. I have also found Ed Norton's critiques of ORs to be persuasive (e.g., doi.org/10.1111/1475-6773.14337 for binary outcomes). Is there a counterargument that you would recommend that I read here? | |
| Aug 9 at 20:47 | comment | added | Frank Harrell | The two give similar fits so I always use the proportional odds model because the effects (odds ratios) are much easier to interpret than the estimated effects in a probit ordinal model. Logit also has some minor computational benefits. | |
| Aug 9 at 20:17 | history | answered | dimitriy | CC BY-SA 4.0 |