Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: Conditional logistic regression within GAM
From
Kieran McCaul <[email protected]>
To
"[email protected]" <[email protected]>
Subject
RE: st: Conditional logistic regression within GAM
Date
Mon, 23 Sep 2013 06:13:48 +0800
...
-mkspline- creates variables that define the spline and these are then added to your model. I haven't used -fp-, but it seems to me that -mkspline- works in a similar fashion to -fp generate-. The strata variables in -clogit- shouldn't be an issue.
Interpreting the coefficients of these spline variables requires a bit of work, but this has been facilitated by the user-written program -xblc- (Orsini and Greenland, 2011) which you should have a look at . Alternatively, -fp- has a series of post estimation commands that aid in the interpretation of the output of these programs.
Orsini N and Greenland S (2011). A procedure to tabulate and plot results after flexible modeling of a quantitative covariate. Stata J 11(1): 1-29.
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Friday, 20 September 2013 1:18 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
One of the major points of the paper is to compare fps to non-parametric methods, where confidence intervals are based on local support, not distance of a covariate from the mean....but I hadn't thought about your mkspline suggestion. In looking at the documentation of mkspline, I didn't see any functionality for defining strata. Do you know of mkspline being used in such a capacity? Thanks again,
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Kieran McCaul
Sent: Thursday, September 19, 2013 1:03 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
...
A case-crossover study: OK, that rules out my previous suggestion.
Is there any particular reason you want to use GAM?
Presumably you want to model the number of drinks as a curvilinear effect, so why not use fractional polynomials (-fp-) or restricted cubic splines (-mkspline-)?
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Thursday, 19 September 2013 2:09 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
Thanks so much for your willingness to respond. My case-control situation is this. The case-control matching is that of 2 responses...the first is a single respondents drinking prior to an injury they experienced....the corresponding matched observation is the drinking for that same individual during the same 6-hour period they were injured on the same day of the week 7-days prior. So the only matching that I have is for the 1-1 matched pairs...there are no other matching criteria (e.g., age, gender) that are being considered as the data is intended to be used in a case-crossover analysis. So the outcome in the conditional logistic regression model is either a 1 or a 0 (corresponding to the injury period or the control period) and the exposure is the number of drinks consumed in the corresponding 6-hour period. So I'm trying to find the analogous method in GAM using conditional logistic regression that is implemented in, for example, -fp-. Thanks again for any insight,
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Kieran McCaul
Sent: Tuesday, September 17, 2013 7:03 PM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
...
OK, Plan B.
How many matching strata do you actually have? Note, the number of matching strata does not equate to the number of matched pairs. If the data are matched on age and sex, for example, and you have a number of matched pairs that are female/aged 65, these should all be treated as one matching strata. Same applies if you have more than one control per case.
If the number of strata are comparatively small, you could use ordinary logistic regression and account for the matching by simply adjusting for the matching variables. This is commonly done when case-control studies employ frequency-matching, for example.
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Wednesday, 18 September 2013 5:31 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
Looking into this further, although Cox proportional hazards modeling can be used to estimate conditional logistic regression models, the exact partial-likelihood method for handling tied failures must be used...an option which looks not to be available in GAM.
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Tuesday, September 17, 2013 1:40 PM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
Great...this helps...is there any reference you know of that describes the implementation of conditional logistic regression models using Cox proportional hazards modeling? Thanks again,
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Kieran McCaul
Sent: Monday, September 16, 2013 11:19 PM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
...
You can fit conditional logistic regression using Cox proportional hazards.
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Tuesday, 17 September 2013 6:57 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
I was wondering if anyone knew of Stat's GAM implementing conditional logistic regression models...I guess I'm a bit surprised that I can't seem to find it given that conditional logistic regression models are one of the examples used in Hastie and Tibshirani's 'Generalized Additive Models' book. Does anyone know of GAM (or any other type of non- or semi-parametric models) that allow the fitting of conditional logistic models in Stata? Thanks much again in advance,
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Jason Bond
Sent: Wednesday, September 11, 2013 9:53 AM
To: [email protected]
Subject: RE: st: Conditional logistic regression within GAM
Actually, the point of a paper I am writing is to compare fractional polynomial approaches with non- and semi-parametric approaches. There are some pretty dramatic differences between the methods (e.g., confidence limits and others) that I'd like to highlight. Thanks,
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Steve Samuels
Sent: Tuesday, September 10, 2013 2:54 PM
To: [email protected]
Subject: Re: st: Conditional logistic regression within GAM
-mfp- and -fp- work with -clogit- and has the same purpose as -gam-
Steve
On Sep 10, 2013, at 5:12 PM, Jason Bond wrote:
I am trying to fit a generalized additive model to a matched case-control dataset to model a dichotomous outcome (i.e., using conditional logistic regression) as a function of a continuous risk factor. In the help for the Royston and Ambler 'gam' ado function, the only options allowed are a binomial family and a logit link...which doesn't estimate conditional logistic models. Does anyone know of an implementation of such a procedure? Thanks much in advance,
Jason
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/