Timeline for Highly significant coefficient does not increase R²
Current License: CC BY-SA 3.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 6, 2013 at 20:04 | comment | added | Jeremy Miles | it's certainly possible. But whether multicollinearity is an issue depends on your data, and the question you're asking. For example, if I've got two tests of math ability, and I want to control for math ability, these two tests will be highly correlated, but I'm still going to put them into the model, because I want to make sure I've controlled for math ability. | |
| Apr 6, 2013 at 13:22 | comment | added | Magnus | thanks for this discussion here as well. Does that mean that (multi)collinearity might be an issue in my model? | |
| Apr 6, 2013 at 0:49 | history | edited | Jeremy Miles | CC BY-SA 3.0 |
More about not necessarily.
|
| Apr 5, 2013 at 23:58 | comment | added | Jeremy Miles | Agreed. I meant not a problem in that this does not mean something has gone wrong. It might be indicative of some other sort of problem, so I've amended it to read "not necessarily a problem". | |
| Apr 5, 2013 at 23:57 | history | edited | Jeremy Miles | CC BY-SA 3.0 |
Added not necessarily a problem.
|
| Apr 5, 2013 at 23:55 | comment | added | whuber♦ | I think this example--although it will likely provide useful insights upon further analysis--may be somewhat misleading, Jeremy. Your conclusion "it's not a problem" actually belies a huge problem: the $R^2$ barely changes even when the new variable is highly significant because the system is very nearly collinear. (The condition number is over $100$, which is enormous for such a small design matrix.) | |
| Apr 5, 2013 at 21:45 | history | answered | Jeremy Miles | CC BY-SA 3.0 |