R help - Apr 2012 - question on jitter in plot.Predict in rms

Dear colleagues,

I have a question regarding controlling the jitter when plotting
predictions in the rms package.  Below I've simulated some data that
reflect what I'm working with.  The model predicts a continuous variable
with an ordinal score, a two-level group, and a continuous covariate.  Of
primary interest is a plot of the group by score interaction, where the
score is the ordinal variable, and the group Ns are quite disparate.

When I produce the plot for the predicted values with the data=llist
argument, as expected I get datadensity hatch marks.  However, in the group
with the smaller N, I get jittered datadensity points, while in the group
with the larger N, the jitter apparently defaults to single vertical lines,
which I assume is because the jitter would look like a black blob.  Some of
my co-authors are a bit worried about how that looks, so here I am.

Apart from abandoning data=llist altogether, is there a way to modify the
jitter in the smaller group so it behaves like the larger one?

Of secondary importance, anything you can tell me about improving my clumsy
little simulation code would be welcome--I have little to no idea what I'm
doing there.  for example, would there be a way to produce the group
variable with the disparate Ns more directly?

Thanks,

Mike Babyak
Behavioral Medicine Research Center
Duke University Medical Center



#question about jitter/llist in rms
#R v 2.14.1 under windows 7
####################################################################

#question about jitter/llist in rms
require(rms)
#simulate some data
n     = 5000
age   = runif(n)
score = runif(n) + 0.5*age
group<- as.numeric(sample(c(FALSE,TRUE), 5000, replace=T, prob=c(.1, .9)))
ordscore = as.numeric(factor(rep(1:5, length.out=n)))
table(group,ordscore)
e     = rnorm(n, 0, 0.1)

#true model
y     = group + 0.6*ordscore + group*ordscore  + .2*age + e

#convert group to factor
group.f<-as.factor(group)

#save data characterics
dd1<-datadist(age,ordscore,group.f)
options(datadist="dd1")

#estimate model
reg1<-ols(y~group.f+ordscore+group.f*ordscore+age,x=T,y=T)

#plot results
preg<-Predict(reg1,ordscore,group.f)

#produces interaction plot with datadensity hatch marks
plot(preg,data=llist(ordscore,group.f))

	[[alternative HTML version deleted]]

Mike,

Try plot(pref, ..., scat1d.opts=list(frac=0.025, lwd=0.3, nhistSpike=i))

where i = 1 to always use spike histograms (default is to use them if n
>2000) or i=1e7 to never use them and to always jitter instead.  There are
many other scat1d options you can pass through scat1d.opts.

Frank



Mike Babyak wrote
> 
> Dear colleagues,
> 
> I have a question regarding controlling the jitter when plotting
> predictions in the rms package.  Below I've simulated some data that
> reflect what I'm working with.  The model predicts a continuous
variable
> with an ordinal score, a two-level group, and a continuous covariate.  Of
> primary interest is a plot of the group by score interaction, where the
> score is the ordinal variable, and the group Ns are quite disparate.
> 
> When I produce the plot for the predicted values with the data=llist
> argument, as expected I get datadensity hatch marks.  However, in the
> group
> with the smaller N, I get jittered datadensity points, while in the group
> with the larger N, the jitter apparently defaults to single vertical
> lines,
> which I assume is because the jitter would look like a black blob.  Some
> of
> my co-authors are a bit worried about how that looks, so here I am.
> 
> Apart from abandoning data=llist altogether, is there a way to modify the
> jitter in the smaller group so it behaves like the larger one?
> 
> Of secondary importance, anything you can tell me about improving my
> clumsy
> little simulation code would be welcome--I have little to no idea what
I'm
> doing there.  for example, would there be a way to produce the group
> variable with the disparate Ns more directly?
> 
> Thanks,
> 
> Mike Babyak
> Behavioral Medicine Research Center
> Duke University Medical Center
> 
> 
> 
> #question about jitter/llist in rms
> #R v 2.14.1 under windows 7
> ####################################################################
> 
> #question about jitter/llist in rms
> require(rms)
> #simulate some data
> n     = 5000
> age   = runif(n)
> score = runif(n) + 0.5*age
> group<- as.numeric(sample(c(FALSE,TRUE), 5000, replace=T, prob=c(.1,
.9)))
> ordscore = as.numeric(factor(rep(1:5, length.out=n)))
> table(group,ordscore)
> e     = rnorm(n, 0, 0.1)
> 
> #true model
> y     = group + 0.6*ordscore + group*ordscore  + .2*age + e
> 
> #convert group to factor
> group.f<-as.factor(group)
> 
> #save data characterics
> dd1<-datadist(age,ordscore,group.f)
> options(datadist="dd1")
> 
> #estimate model
> reg1<-ols(y~group.f+ordscore+group.f*ordscore+age,x=T,y=T)
> 
> #plot results
> preg<-Predict(reg1,ordscore,group.f)
> 
> #produces interaction plot with datadensity hatch marks
> plot(preg,data=llist(ordscore,group.f))
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help@ mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context:
http://r.789695.n4.nabble.com/question-on-jitter-in-plot-Predict-in-rms-tp4598555p4600795.html
Sent from the R help mailing list archive at Nabble.com.