Date:         Mon, 11 Mar 2002 09:56:17 -0600
Reply-To:     Alex Shackman <shackman@psyphw.psych.wisc.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Alex Shackman <shackman@psyphw.psych.wisc.edu>
Subject:      [kkhoie@USA.COM: Normalizing skewed distributions using SPSS 6.0]
Content-Type: text/plain; charset=us-ascii

The appropriate transformation will depend upon the magnitude of the skew.

In order of "strength" (a la J. Tukey), from minimal transforming to
maximal:

(x)^.75
(x)^.66 {2/3 root}
sqrt(x) = (x)^.5
(x)^.33 {cube root}
(x)^.25
log10(x) = ln(x)
1/x {reciprocal transformation}

Occasionally, more exotic transforms are useful. For example,
1/(sqrt(x)) or log(log(x)).

NB: you may need to add a constant to x to maintain the appropriate
rank ordering of cell means, as when some observations are negatively
signed.

**Always check the rank order of cell means after transformation**

To choose the "optimal" transformation, the skewness, kurtosis and
homogeneity of variances should be compared.

Best,
Alex Shackman

----- Forwarded message from Kathy Khoie <kkhoie@USA.COM> -----

Newsgroups:   bit.listserv.spssx-l
Date:         Mon, 11 Mar 2002 02:30:43 -0500
Reply-To: Kathy Khoie <kkhoie@USA.COM>
From: Kathy Khoie <kkhoie@USA.COM>
Subject:      Normalizing skewed distributions using SPSS 6.0
To: SPSSX-L@LISTSERV.UGA.EDU

Hi,
I'm new to the list. Could not find a FAQ or archive. Need some input on
normalizing skewed distributions. Any guidance on this topic will be
appreciated,

Regards,
Kathy

----- End forwarded message -----