| |
 |
|
|
Science Forum Index » Statistics - Math Forum » transforming ratio scores
Page 1 of 1
|
| Author |
Message |
| Geoff Hammond |
 Posted: Wed Jan 03, 2007 9:15 pm |
|
|
|
Guest
|
Raio scores are common in some areas of physiological research. Whenever I
use them, I routinely take a log transform before analysis, and back
transform for presentation. A reviewer of a manuscript has objected to
this practice, and says that the transform should not be done unless there
is evidence on non-normality in the data. What are the views of the
group -- transform routinely, or not?
Geoff Hammond |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Thu Jan 04, 2007 2:27 am |
|
|
|
Guest
|
On Thu, 4 Jan 2007 10:15:50 +0900, "Geoff Hammond"
<geoff@psy.uwa.edu.au> wrote:
Quote: Raio scores are common in some areas of physiological research. Whenever I
use them, I routinely take a log transform before analysis, and back
transform for presentation. A reviewer of a manuscript has objected to
this practice, and says that the transform should not be done unless there
is evidence on non-normality in the data. What are the views of the
group -- transform routinely, or not?
Three main reasons for transforming data are
1) for achieving "equal intervals" in some hypothesized
metric (for example, so that linearity is possible0;
2) for achieving homogeneity of measurement variance (or
prediction variance, which is not entirely the same thing but
similar) across its range;
3) for achieving normality.
Lack of normality is probably the easiest indicator that
something may be wrong, but it is a less important reason
to do something, compared to the other two.
I do automatically *consider* transforming ratios by taking
logs, or transforming counts by taking square roots, but my
default or 'routine' is to leave them in the units that folks
are accustomed to. Less confusion, less to explain.
--
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| Ray Koopman |
| Posted: Thu Jan 04, 2007 4:10 am |
|
|
|
Guest
|
Geoff Hammond wrote:
Quote: Raio scores are common in some areas of physiological research. Whenever I
use them, I routinely take a log transform before analysis, and back
transform for presentation. A reviewer of a manuscript has objected to
this practice, and says that the transform should not be done unless there
is evidence on non-normality in the data. What are the views of the
group -- transform routinely, or not?
Geoff Hammond
The reviewer is wrong. Normality is the last thing you should worry
about. Rich's hierarchy is OK, provided the form of the model is
correct -- you wouldn't want to transform to fit to a wrong model. |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Fri Jul 25, 2008 6:15 pm
|
|
