On 13/11/2014 11:08, Marc Girondot wrote:
Le 13/11/2014 01:26, MacQueen, Don a écrit :
Along the lines of what Bert Gunter said, the ideal way to represent <LDL
results depends on the functions used later to analyze them. I deal with
such data on a daily basis and have never found it necessary to
incorporate that information in the same variable as the results. What
would you do if data were censored at both ends, both low and high?
Anyway, the functions I use mostly incorporate that information in a
second variable, a ³detection indicator² variable, and that¹s what I do.
-Don
I agree that LDL is a special case of what could be named ODL (Out of
detection limit).
To answer to Bert Gunter, indeed if LDL (or ODL) values are changed into
NA, the results will be biased. That's why I would like to introduce
another category. I don't plan to just transform them as NA.
But thinking again about this problem, a LDL must be always associated
with one value (or two in the case of ODL) that indicates the detection
limit. In a dataset, all values have not necessarily the same limit
depending on the experimental conditions.
The best solution that I find is to use attributes to indicate the
limits. A NA attribute for a NA value will be treated as a "true" NA.
For exemple:
> values <- c(NA, 29, 30, NA, 3)
> attributes(values) <- list(ODL=c(NA, "[10, 40]", "[0, 40]", "[0,
40]", "[0, 40]"))
> values
[1] NA 29 30 NA 3
attr(,"ODL")
[1] NA "[10, 40]" "[0, 40]" "[0, 40]" "[0, 40]"
> values[3]
[1] 30
> attributes(values)$ODL[3]
[1] "[0, 40]"
> values[1]
[1] NA
> attributes(values)$ODL[1]
[1] NA
The attributes are retained in data.frame. So it seems to be a good
solution.
> essai <- data.frame(c1=values)
> essai
c1
1 NA
2 29
3 30
4 NA
5 3
> essai$c1
[1] NA 29 30 NA 3
attr(,"ODL")
[1] NA "[10, 40]" "[0, 40]" "[0, 40]" "[0, 40]"
Thanks to the list members,
Marc
I strongly recommend you re-read and take action on Bert Gunter's
comment, quoted here for truth!
-----------------------------
Ouch!
The values are **NOT** missing -- they are (left) censored, and need
to be handled by appropriate censored data methods. I suggest you
(all!) either read up on this or consult someone locally who has
knowledge of such methods.
-- Bert
------------------------
You are re-inventing the wheel and yours will probably end up square!
R already has facilities for handling censored data, e.g. Surv in the
survival package (which despite its name is applicable to applications
other than survival analysis).
______________________________________________
R-help@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.
