Timeline for Precisely how does R's coxph() handle repeated measures?
Current License: CC BY-SA 3.0
16 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 11, 2020 at 14:32 | history | edited | CommunityBot |
Commonmark migration
|
|
| Nov 9, 2015 at 21:17 | comment | added | user76943 | Thanks for the edits, Theodor. Your comments and points have provided excellent guidance to the subject | |
| Nov 9, 2015 at 8:36 | history | edited | Theodor | CC BY-SA 3.0 |
improved clarity according to suggestions
|
| Nov 8, 2015 at 17:15 | comment | added | user76943 | Ah yes, it is Efron by default. I was confused since you reference section 8.2 in (1) above in Therneau & Grambsch's book, where in the 4th paragraph of 8.2.1 they say "We have purposefully used the Breslow approximation...to avoid for any correction for ties; doing this leads to identical coefficients..." It would be nice, but not necessary, to gain clarity as to what you're referencing (statistically) in the cited reference that supports your assertion in (1). | |
| Nov 8, 2015 at 10:24 | comment | added | Theodor | I did not think about ties in the answer above. coxph() does not use the Breslow as default, it should be specified as method = "breslow" in the argument. | |
| Nov 8, 2015 at 3:41 | comment | added | user76943 | Also note: as explained in Therneau and Grambsch's book, the reason the answer in (1) above is correct is because coxph() uses the Breslow method as default for ties. | |
| Nov 2, 2015 at 9:53 | history | edited | Theodor | CC BY-SA 3.0 |
deleted 19 characters in body
|
| Nov 2, 2015 at 2:39 | comment | added | user76943 | The reference link you provided in (1) should be: springer.com/us/book/9780387987842 (assuming you're citing Therneau and Grambsch's book) | |
| Oct 31, 2015 at 15:41 | vote | accept | CommunityBot | ||
| Oct 31, 2015 at 15:41 | comment | added | user76943 | (2) may be rephrased as: if cluster(ID) is used, a "robust" estimate of standard errors is imposed and possible dependence between subjects is measured (e.g. by standard errors and variance scores). Not using cluster(ID), on the other hand, imposes independence on each observation and more "information" is assumed in the data. | |
| S Oct 30, 2015 at 14:05 | history | edited | gung - Reinstate Monica | CC BY-SA 3.0 |
formatted
|
| Oct 30, 2015 at 13:48 | review | Suggested edits | |||
| S Oct 30, 2015 at 14:05 | |||||
| Oct 29, 2015 at 18:49 | review | Suggested edits | |||
| Oct 29, 2015 at 19:06 | |||||
| Oct 29, 2015 at 15:56 | comment | added | Theodor | I meant: if the observations are clustered, then they might or might not be independent. Assuming that they are independent (i.e. not using cluster(id) ) is almost certainly wrong in this case, but there is no idea of knowing that beforehand | |
| Oct 28, 2015 at 16:20 | comment | added | user76943 | Thank you. Regarding (2): can "This is because if you (wrongly) assume..." be replaced by "If you do not use cluster(ID) in the call to coxph(), then you wrongly assume...." | |
| Oct 28, 2015 at 15:52 | history | answered | Theodor | CC BY-SA 3.0 |