Normalizing data
nos...@my-deja.com
distribution)? If so, how do I begin? I am using SPSS 10.0.
Sent via Deja.com http://www.deja.com/
Before you buy.
Jonathan Fry
In the menus, Transform/Rank Cases/Rank Types/More>>/Normal Scores.
Jonathan Fry
SPSS Inc.
SPSS questions to sup...@spss.com
Alan Acock
is not to get normal scores (z-scores) but to transform a the variable into
a normal distribution. PRELIS (Part of LISREL) can do this, but I don't
think SPSS can--would love to be wrong on this.
Alan Acock
--
***************************************************
The Acock's
Alan Acock's Address is ac...@home.com
Toni Acock's is antoni...@home.com
"Jonathan Fry" <j...@spss.com> wrote in message
news:38A1C421...@spss.com...
Peter Gallagher
shape of the distribution, I've often found that a simple log10 or square root
transformation using the Compute function is enough. The real problem is if
there are a large proportion of values around 0 - if this is the case, then I
haven't yet found a reasonable transformation to correct this.
nos...@my-deja.com
did go through the steps Jonathan suggested, and I did manage to obtain
a new set of scores with a mean of 0 and a standard deviation of 1, but
the distribution (shape) of the scores was still the same.
Assuming that you are correct and SPSS doesn't provide this feature, I
would like to suggest that some of the SPSS screen displays and names of
some of the generated variables be modified to use the word/label
standardized instead of normalized. It might avoid some confusion for
other users.
In article <LAro4.20508$_G1.3...@news1.sttls1.wa.home.com>,
Rich Ulrich
> Yes, Alan, I want to transform the scores into a normal distribution. I
> did go through the steps Jonathan suggested, and I did manage to obtain
> a new set of scores with a mean of 0 and a standard deviation of 1, but
> the distribution (shape) of the scores was still the same.
>
> Assuming that you are correct and SPSS doesn't provide this feature, I
> would like to suggest that some of the SPSS screen displays and names of
> some of the generated variables be modified to use the word/label
> standardized instead of normalized. It might avoid some confusion for
> other users.
< snip, someone else: PRELIS which accompanies LISREL offers a
transformation to the Gaussian shape.>
Well, I don't think that "normalize" ought to confuse users, much more
or less than "standardize." My first guess for either is that it
would have to be a linear transformation. And you may be wandering
around in a pretty dense fog, so you have probably gotten lost at 3 or
4 other turns in the analyses -- if you were regularly taking so much
on faith.
Maybe that is stating it too strongly. But I have hardly seen anyone
transform to normal shape, and I have gotten just about ZERO response
in these stat-groups when I have mentioned that prospect (such as, for
a nice version of a robust, non-parametric t-test).
Is this an example of a general phenomenon? Folks are apt to be
confused when they have used just ONE computer program for Task-A, and
assume that it must establish (or follow) the world's conventions. I
tend to get that way about SPSS, myself.
Since I have read discussions about "What is the formula for
*percentile*", I think that if you are going to do any formal
analysis that is based on a normal shape, and you expect that someone
else might replicate, then you *ought* to state explicitly what
add-on fraction you are using, before computing the inverse-normal.
Or what other way you are finding your value. And your software
surely ought to tell you, so you can report it.
If PRELIS is not highly explicit to that extent about what it is
doing, then -- are you *sure* you know what it is doing?
--
Rich Ulrich, wpi...@pitt.edu
http://www.pitt.edu/~wpilib/index.html
Jonathan Fry
>
> On Thu, 10 Feb 2000 15:45:02 GMT, nos...@my-deja.com wrote:
>
> > Yes, Alan, I want to transform the scores into a normal distribution. I
> > did go through the steps Jonathan suggested, and I did manage to obtain
> > a new set of scores with a mean of 0 and a standard deviation of 1, but
> > the distribution (shape) of the scores was still the same.
> >
> > Assuming that you are correct and SPSS doesn't provide this feature, I
> > would like to suggest that some of the SPSS screen displays and names of
> > some of the generated variables be modified to use the word/label
> > standardized instead of normalized. It might avoid some confusion for
> > other users.
My apologies to Rich Ulrich for using his post to reply to Nosweat's
post. My news reader has not seen Nosweat's post yet.
When I try the steps I outlined in my earlier post, the resulting
variable has a nice, bell-shaped histogram. The "normal scores" are
given a normal distribution. The normal scores are computed by inverse
normal transforms on proportion estimates. SPSS offers a choice of
documented methods for the proportion estimates, and includes the choice
made in the variable label of the generated variable.
Jonathan Fry
SPSS Inc.
gftem...@hotmail.com
> Is it possible to normalize a variable (i.e. transform it into a normal
> distribution)? If so, how do I begin? I am using SPSS 10.0.
>
>
> Sent via Deja.com http://www.deja.com/
> Before you buy.
2011 Templeton, G.F. "A Two-Step Approach for Transforming Continuous Variables to Normal: Implications and Recommendations for IS Research," Communications of the AIS, Vol. 28, Article 4. Available online.
The procedures above will help you retain the original mean and variance, or standardize by using mean = 0 and variance = 1.