Timeline for Coefficient of determination for binary responses
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 22, 2013 at 14:53 | vote | accept | Cesare Camestre | ||
| Aug 22, 2013 at 14:49 | comment | added | Nick Cox | I don't know what "analogy" you have in mind. The point is that there is no such analogy. If you are guessing that the Cox-Wermuth limit applies to linear regression with a continuous response and binary predictors, then that's not so. I've already given a counter-example. | |
| Aug 22, 2013 at 14:44 | comment | added | Cesare Camestre | To cut it short, its seems that this analogy applies only for dependent variables. If the dependent variable is 1, or 0. If the control variables are binary variables this shouldn't affect the value of the R square. Correct? | |
| Aug 22, 2013 at 14:39 | comment | added | Nick Cox | Responses are not predictors. | |
| Aug 22, 2013 at 14:31 | comment | added | Cesare Camestre | They state clearly "the interpretation [of the R squared] is misleading in linear regressions with binary responses since low value of R^2, roughly .1 are inevitable even if an important relation is present". Yet you seem to suggest that this can be one. What am I getting wrong here. | |
| Aug 22, 2013 at 14:26 | comment | added | Nick Cox | Yes. I've not read the Cox and Wermuth paper for some years, but I don't think they said or implied that. | |
| Aug 22, 2013 at 14:22 | comment | added | Cesare Camestre | So if I am understanding well, if I have one continuous predictor, and 5 binary predictors, it is wrong to assume that "that the maximum value the R squared can be is 0.36". | |
| Aug 22, 2013 at 13:57 | history | answered | Nick Cox | CC BY-SA 3.0 |