A lot depends on what you are trying to do with the factor analysis.
Create a scale (index)? etc. The subject matter matters. How many
variables are there? How many cases?
Are the variables selected to represent different domains or constructs?
One hopes there will be a few factors that account for much of the
variance and many that are not needed.
By tradition one does not even extract factors that do not even account
for one variable's worth of the total variance accounted for. This is a
rule that says there is just no way more factors could be useful. It
does not speak to the number to retain.After all, in general factor,
analysis is done to represent many variables in as few as makes sense in
the circumstance. However, this is just a rule to ease the computer
burden. The number to retain in a final solution is a much smaller
number of factors. Parallel analysis (search the archives of this list
for syntax) is sometimes useful. I some times use a guess for the max
possible number of factors to retain 1 variable's worth (eigenvalue of
1.00) more than what would be found in random data with the given number
of cases and variables.
Art Kendall
Social Research Consultants
Mark Webb wrote:
> Are there any rule-of-thumb guidlines regarding the amount of variance
> explained in factor analysis solutions.
> [Using >1 Eiganvalue, varimax rotation, principal components,etc] -
> nothing fancy.
>
> Does one accept above a certain % reject below some %?
> Is a low % indicative of a weak model ?
> Should one even look at this statistic ?
>
> --
> Mark Webb
>
> +27 21 786 4379
> +27 72 199 1000
> Skype - webbmark
>
[hidden email]
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
>
[hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Art Kendall
Social Research Consultants
Locked
