R help - Feb 2010 - Ordering categories on a boxplot - a serious trap??

Hello all,

I think I probably did something stupid, and R's part was to allow me to do
it.  My goal was to control the order of factor levels appearing horizontally on
a boxplot.  Enter search engines and perhaps some creative stupidity on my part,
and I came up with the following:

	v=read.table("factor-order.txt",header=TRUE);
	levels(v$doseGroup) = c("L", "M", "H");
	boxplot(v$dose~v$doseGroup);


A good way to see the trap is to evaluate:

	v=read.table("factor-order.txt",header=TRUE);
	par(mfrow=c(2,1));
	boxplot(v$dose~v$doseGroup);
	levels(v$doseGroup) = c("L", "M", "H");
	boxplot(v$dose~v$doseGroup);
	par(mfrow=c(1,1));

The above creates two plots, one correct with the factors in an inconvient
order, and one that is WRONG.  In the latter, the labels appear in the desired
order, but the data does not "move with them."  I did not discover the
problem until I repeated the same type of plot with something that had a known
relationship with the levels, and the result was clearly not correct.

I *think* the problem is to assign to the return value of levels().  How did I
think to do that?  I'm not really sure, but please look at

  https://stat.ethz.ch/pipermail/r-help/2008-August/171884.html


Perhaps it does not say to do exactly what I did, but it sure was easy to follow
to the mistake, it appeared to do what I wanted, and the consequences of the
mistake are ugly.  Perhaps levels() should return something that is immutable?? 
If I am looking at this correctly, levels() is an accident waiting to happen.

What should I have done?  It seems:

	read data and order factor levels
	v=read.table("factor-order.txt",header=TRUE);
	group = factor(v$doseGroup,levels = c("L", "M",
"H") );
	boxplot(v$dose~group);


One disappointment is that the above factor() call apparently needs to be
repeated for any subset of v - I'm still trying to get my mind around that
one.

Can anyone confirm this?  It strikes me as a trap that should be addressed so
that an error results rather than a garbage graph.

Bill


---
Wilhelm K. Schwab, Ph.D.

> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Schwab,Wilhelm K
> Sent: Thursday, February 25, 2010 3:51 PM
> To: r-help at r-project.org
> Subject: [R] Ordering categories on a boxplot - a serious trap??
> 
> Hello all,
> 
> I think I probably did something stupid, and R's part was to 
> allow me to do it.  My goal was to control the order of 
> factor levels appearing horizontally on a boxplot.  Enter 
> search engines and perhaps some creative stupidity on my 
> part, and I came up with the following:
> 
> 	v=read.table("factor-order.txt",header=TRUE);
> 	levels(v$doseGroup) = c("L", "M", "H");
> 	boxplot(v$dose~v$doseGroup);
levels<- translated the current level labels into
another language, it did not change the integer
codes of the factor.  If you want to reorder the
levels call factor(..., levels=).  E.g.,

  > z <-
factor(c("Small","Large","Medium","Small"))
  > str(z)
   Factor w/ 3 levels "Large","Medium",..: 3 1 2 3
  > str(factor(z,
levels=c("Small","Medium","Large")))
   Factor w/ 3 levels "Small","Medium",..: 1 3 2 1

You can relabel them also by using the labels= argument
to factor
  > str(factor(z,
levels=c("Small","Medium","Large"),
labels=c("S","M","L")))
 Factor w/ 3 levels "S","M","L": 1 3 2 1

Calling levels<- changes nothing but the level labels:
  > zcopy <- z
  > levels(zcopy) <-
c("Small","Medium","Large")
  > str(zcopy)
    Factor w/ 3 levels "Small","Medium",..: 3 1 2 3

levels<- is handy for low-level manipulations but not
for general use.  Even factor(,levels=) can be a bit
dangerous: if a new level is misspelled it will silently
add NA's to the data:
  > str(factor(z, levels=c("Smal", "Medium",
"Large")))
   Factor w/ 3 levels "Smal","Medium",..: NA 3 2 NA

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com 
> 
> 
> A good way to see the trap is to evaluate:
> 
> 	v=read.table("factor-order.txt",header=TRUE);
> 	par(mfrow=c(2,1));
> 	boxplot(v$dose~v$doseGroup);
> 	levels(v$doseGroup) = c("L", "M", "H");
> 	boxplot(v$dose~v$doseGroup);
> 	par(mfrow=c(1,1));
> 
> The above creates two plots, one correct with the factors in 
> an inconvient order, and one that is WRONG.  In the latter, 
> the labels appear in the desired order, but the data does not 
> "move with them."  I did not discover the problem until I 
> repeated the same type of plot with something that had a 
> known relationship with the levels, and the result was 
> clearly not correct.
> 
> I *think* the problem is to assign to the return value of 
> levels().  How did I think to do that?  I'm not really sure, 
> but please look at
> 
>   https://stat.ethz.ch/pipermail/r-help/2008-August/171884.html
> 
> 
> Perhaps it does not say to do exactly what I did, but it sure 
> was easy to follow to the mistake, it appeared to do what I 
> wanted, and the consequences of the mistake are ugly.  
> Perhaps levels() should return something that is immutable??  
> If I am looking at this correctly, levels() is an accident 
> waiting to happen.
> 
> What should I have done?  It seems:
> 
> 	read data and order factor levels
> 	v=read.table("factor-order.txt",header=TRUE);
> 	group = factor(v$doseGroup,levels = c("L", "M",
"H") );
> 	boxplot(v$dose~group);
> 
> 
> One disappointment is that the above factor() call apparently 
> needs to be repeated for any subset of v - I'm still trying 
> to get my mind around that one.
> 
> Can anyone confirm this?  It strikes me as a trap that should 
> be addressed so that an error results rather than a garbage graph.
> 
> Bill
> 
> 
> ---
> Wilhelm K. Schwab, Ph.D.
> 
> ______________________________________________
> R-help at r-project.org 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.
>

Phil,

That works[*], but I still think there is a big problem given how easy it is to
do the wrong thing, and that searches lead to dangerous instructions.  Hopefully
this will serve to keep others out of trouble, but so might an immutable return
value from levels().

[*] I have not yet done anything with selecting parts of the data frame.  Using
a separate factor, I quickly hit trouble with size mismatches, though I could
probably work around them by recreating the factor after any such change. 
Proceeding with caution...

Bill

---
Wilhelm K. Schwab, Ph.D.



-----Original Message-----
From: Phil Spector [mailto:spector at stat.berkeley.edu]
Sent: Thursday, February 25, 2010 7:06 PM
To: Schwab,Wilhelm K
Subject: Re: [R] Ordering categories on a boxplot - a serious trap??

Wilhelm -
    I don't know if this is correct for your problem because you didn't
provide a reproducible example, but perhaps you could try

v$doseGroup = factor(v$doseGroup,levels=c("L", "M",
"H"))

instead of setting the levels directly.

 					- Phil Spector
 					 Statistical Computing Facility
 					 Department of Statistics
 					 UC Berkeley
 					 spector at stat.berkeley.edu

On Thu, 25 Feb 2010, Schwab,Wilhelm K wrote:
> Hello all,
>
> I think I probably did something stupid, and R's part was to allow me
to do it.  My goal was to control the order of factor levels appearing
horizontally on a boxplot.  Enter search engines and perhaps some creative
stupidity on my part, and I came up with the following:
>
> 	v=read.table("factor-order.txt",header=TRUE);
> 	levels(v$doseGroup) = c("L", "M", "H");
> 	boxplot(v$dose~v$doseGroup);
>
>
> A good way to see the trap is to evaluate:
>
> 	v=read.table("factor-order.txt",header=TRUE);
> 	par(mfrow=c(2,1));
> 	boxplot(v$dose~v$doseGroup);
> 	levels(v$doseGroup) = c("L", "M", "H");
> 	boxplot(v$dose~v$doseGroup);
> 	par(mfrow=c(1,1));
>
> The above creates two plots, one correct with the factors in an inconvient
order, and one that is WRONG.  In the latter, the labels appear in the desired
order, but the data does not "move with them."  I did not discover the
problem until I repeated the same type of plot with something that had a known
relationship with the levels, and the result was clearly not correct.
>
> I *think* the problem is to assign to the return value of levels().  
> How did I think to do that?  I'm not really sure, but please look at
>
>  https://stat.ethz.ch/pipermail/r-help/2008-August/171884.html
>
>
> Perhaps it does not say to do exactly what I did, but it sure was easy to
follow to the mistake, it appeared to do what I wanted, and the consequences of
the mistake are ugly.  Perhaps levels() should return something that is
immutable??  If I am looking at this correctly, levels() is an accident waiting
to happen.
>
> What should I have done?  It seems:
>
> 	read data and order factor levels
> 	v=read.table("factor-order.txt",header=TRUE);
> 	group = factor(v$doseGroup,levels = c("L", "M",
"H") );
> 	boxplot(v$dose~group);
>
>
> One disappointment is that the above factor() call apparently needs to be
repeated for any subset of v - I'm still trying to get my mind around that
one.
>
> Can anyone confirm this?  It strikes me as a trap that should be addressed
so that an error results rather than a garbage graph.
>
> Bill
>
>
> ---
> Wilhelm K. Schwab, Ph.D.
>
> ______________________________________________
> R-help at r-project.org 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.
>