Hi All,

I seen over the web on numerous methodologies to calculate relative
importance of predictors in a model.
Some of them

 - Johnsons Relative Importance
 - Kruskal's/Theils
 - Dominance Analysis

I wonder which one of these is the most preferred and accurate methods
and how wwe could use them in SPSS.
Your sincere comments will be fo great help.

Many Thanks.

On Jan 25, 7:20 am, Rockstar <rocks...@googlemail.com> wrote:
> Hi All,
>
> I seen over the web on numerous methodologies to calculate relative
> importance of predictors in a model.
> Some of them
>
>  - Johnsons Relative Importance
>  - Kruskal's/Theils
>  - Dominance Analysis
>
> I wonder which one of these is the most preferred and accurate methods
> and how wwe could use them in SPSS.
> Your sincere comments will be fo great help.
>
> Many Thanks.

See <The "most important variable" ??> at
http://groups.google.ca/group/MedStats/browse_frm/thread/bf8d801d13d22f71

How timely! I find myself asking the exact same question. I am specifically searching the web for "how to" informaton on Johnson's Relative Importance Method as I have been recommended to use this in my dissertation. I haven't finished looking yet, so if I find the answers, I'll try to come back and post. Please do the same! 

Jackpot!

See http://www1.psych.purdue.edu/~jlebreto/relative.htm

On Mon, 25 Jan 2010 15:32:19 -0800 (PST), Ray Koopman <koopman@sfu.ca>
wrote:

>On Jan 25, 7:20 am, Rockstar <rocks...@googlemail.com> wrote:
>> Hi All,
>>
>> I seen over the web on numerous methodologies to calculate relative
>> importance of predictors in a model.
>> Some of them
>>
>>  - Johnsons Relative Importance
>>  - Kruskal's/Theils
>>  - Dominance Analysis
>>
>> I wonder which one of these is the most preferred and accurate methods
>> and how wwe could use them in SPSS.
>> Your sincere comments will be fo great help.
>>
>> Many Thanks.
>
>See <The "most important variable" ??> at
>http://groups.google.ca/group/MedStats/browse_frm/thread/bf8d801d13d22f71


That has a good summary by Ray, concerning many of
the problems.

It does not mention, explicitly, the problems created by
the presence of "suppressor variables" - whether or not
they do deserve to be in any equation (in versus out
create different logical problems).  Or, the notion that 
the correlations, etc., in an equation inevitably limit
you to statements about the *measured*  variable, and
not about the underlying theoretical construct - When is
it appropriate to correct for "attenuation" from errors
of measurement?

-- 
Rich Ulrich