Zhang 2011 (re)analysis
On 11-10-31 05:22 AM, Reinhold Kliegl wrote:
One problem appears to be that 111 id's are renumbered from 1 to 55 (56) in the two groups. Unfortunately, it also appears that there is no unique mapping to treatment groups. So there are some subjects with 8 values assigned to one of the groups.
Thanks. It looks like IDs are nested within center (not within treatment). That doesn't seem to change the story very much (as far as , though (Zhang et al don't report estimated random-effect variances ...)
library(geepack) data(respiratory) resp1 <- respiratory resp1 <- transform(resp1,
+ center=factor(center), + id=factor(id))
str(resp1)
'data.frame': 444 obs. of 8 variables: $ center : Factor w/ 2 levels "1","2": 1 1 1 1 1 1 1 1 1 1 ... $ id : Factor w/ 56 levels "1","2","3","4",..: 1 1 1 1 2 2 2 2 3 3 ... $ treat : Factor w/ 2 levels "A","P": 2 2 2 2 2 2 2 2 1 1 ... $ sex : Factor w/ 2 levels "F","M": 2 2 2 2 2 2 2 2 2 2 ... $ age : int 46 46 46 46 28 28 28 28 23 23 ... $ baseline: int 0 0 0 0 0 0 0 0 1 1 ... $ visit : int 1 2 3 4 1 2 3 4 1 2 ... $ outcome : int 0 0 0 0 0 0 0 0 1 1 ...
detach("package:geepack") ## allow detaching of doBy
detach("package:doBy") ## allow detaching of lme4
The data appear also in the HSAUR package, here the 111 subjects identified with 5 months (visits) each. I suspect month 0 was used as baseline.
library(HSAUR) data(respiratory) resp2 <- respiratory str(resp2)
'data.frame': 555 obs. of 7 variables: $ centre : Factor w/ 2 levels "1","2": 1 1 1 1 1 1 1 1 1 1 ... $ treatment: Factor w/ 2 levels "placebo","treatment": 1 1 1 1 1 1 1 1 1 1 ... $ sex : Factor w/ 2 levels "female","male": 1 1 1 1 1 1 1 1 1 1 ... $ age : num 46 46 46 46 46 28 28 28 28 28 ... $ status : Factor w/ 2 levels "poor","good": 1 1 1 1 1 1 1 1 1 1 ... $ month : Ord.factor w/ 5 levels "0"<"1"<"2"<"3"<..: 1 2 3 4 5 1 2 3 4 5 ... $ subject : Factor w/ 111 levels "1","2","3","4",..: 1 1 1 1 1 2 2 2 2 2 ... Reinhold On Sun, Oct 30, 2011 at 10:00 PM, Ben Bolker <bbolker at gmail.com> wrote:
There's a fairly recent paper by Zhang et al (2011) of interest to folks on this list DOI: 10.1002/sim.4265 In response to a post on the AD Model Builder users' list, I took a quick shot at re-doing some of their results (they have extensive simulation results, which I haven't tried to replicate yet, and an analysis of binary data from Davis (1991) which is included (I *think* it's the same data set -- the description and size of the data set match exactly) in the geepack data set). If anyone's interested, my results so far are posted at http://glmm.wikidot.com/local--files/examples/Zhang_reanalysis.Rnw http://glmm.wikidot.com/local--files/examples/Zhang_reanalysis.pdf So far the R approaches I've tried agree closely with each other and with glmmADMB (except MASS::glmmPQL, which I expected to be different -- the rest all use either Laplace approx. or AGHQ). They *don't* agree with the results Zhang et al got, yet -- I'm sure there's something I'm missing in the contrasts or otherwise ... Suggestions or improvements are welcome. cheers Ben Bolker
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models