factorial analysis results by type of variables

[ro]

Hi all

I have a database with economical and social data (200,000 cases), this data is divided by economic, educational and health variables, for each group of variables included 15-30 variables.

I would like make a factorial analysis and get results by groups of variables ( by economic, educational and health), but i dont know how make the groups and analysis each one in a same analysis...
Its possible??

I grateful any comment

Thanks in advance.
Ro

[Maguin, Eugene]

Ok, so  a case looks something like this
Id econ1 to econ19 educ1 to educ23 health1 to health29

Let's assume all the variables are continuous. So, you can do three factor analyses, one for each variable set (econ, educ, and health). So, suppose you had the results of those analyses. What do you want to do next and why?

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of ro
Sent: Monday, July 15, 2013 5:44 PM
To: [hidden email]
Subject: factorial analysis results by type of variables

Hi all

I have a database with economical and social data (200,000 cases), this data is divided by economic, educational and health variables, for each group of variables included 15-30 variables.

I would like make a factorial analysis and get results by groups of variables ( by economic, educational and health), but i dont know how make the groups and analysis each one in a same analysis...
Its possible??

I grateful any comment

Thanks in advance.
Ro



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/factorial-analysis-results-by-type-of-variables-tp5721185.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

[ro]

Hi Gene,

Thanks for you reply,

I wanna make a index with these results that included this 3 areas. For this, I will use the results of score coefficient matrix.

Regards
Ro

[Art Kendall]

I doubt that a single variable/score/index would be the the most useful summarization of this data.
I do not have access to my files at the moment so do not have the citations from when I did this kind of work at the US Census Bureau.

I suggest you first clarify what you want to do.
Do you want a few score variables that could summarize all of the info across the 3 batteries/domains? I.e., a score could include items from more than one domain? 
Do you want separate score variables for each battery domain?

What do you want to do after you have the scores? predict some variables from outside those you used to get the scores? find clusters?


What constitutes a case?
How were the cases chosen?
How much missing data do you have?
Do you have a unique ID for each case?
Then  re-post to this list.

shooting from the hip, based on what is done in social indicators work, I suggest
1) for each battery separately
do a factor analysis with principal axes and  varimax rotation.
do a parallel analysis say 10,000 times and overlay the eigenvalues from permuted (randomized) data on the scree plot of obtained eigenvalues.
Pick a few candidate numbers of factors to retain using the factor number where the obtained eigen value is at least 1.00 greater than the eigenvalues from the randomized data.
for each candidate number of factors, develop a scoring key using only items that load at least .5 or so on the highest factor and not more than .3 or so on any other factor.
Based on reasonableness of interpretation pick a number of factors to retain. You will end up dropping some items, some will split across factors, and some will simply not load.   You won't necessarily have as many scores as originally retained factors. Some factors may not have enough items to make a score. Some factors may not make any sense.
standardize all of the items that go into scores.
apply the scoring key, i.e. get a mean of all the standardized items that went together on a factor add the computed scores to the data file.
2) do the same thing on as above but treating the 3 batteries as one set.

Now decide what set of scores/indices makes the most sense in the context you intend to use them.

-----------------

if you wanted to find composite scores across all 3 batteries think about In the late 60's or early 70 Doug Carroll from Bell Labs published about n-battery canonical correlation which found dimensions common  across more than the usual 2 batteries for canonical correlation.  This was also a generalization of the multitrait multimethod matrix kind of research. Although Doug passed away several years ago several years ago, several of his students and people developed methods from this approach are on the latter 2 discussion lists.

IIRC n-battery canonical correlation was a grandfather to dual scaling, correspondence analysis, SEM, etc.

post to
The Classification Society List
http://lists.sunysb.edu/index.cgi?A0=CLASS-L

and the Mathematical Psych list
[hidden email]


On 7/16/2013 10:55 AM, ro [via SPSSX Discussion] wrote:

Hi Gene,

Thanks for you reply,

I wanna make a index with these results that included this 3 areas. For this, I will use the results of score coefficient matrix.

Regards
Ro

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/factorial-analysis-results-by-type-of-variables-tp5721185p5721199.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

[ro]

Hi Art,

Thanks for you reply.

I will take your advice and focus in a better way my analysis,
probably I did miss in that point...

Thanks
Regards
Ro




2013/7/17 Art Kendall [via SPSSX Discussion]
<[hidden email]>:

> I doubt that a single variable/score/index would be the the most useful
> summarization of this data.
> I do not have access to my files at the moment so do not have the citations
> from when I did this kind of work at the US Census Bureau.
>
> I suggest you first clarify what you want to do.
> Do you want a few score variables that could summarize all of the info
> across the 3 batteries/domains? I.e., a score could include items from more
> than one domain?
> Do you want separate score variables for each battery domain?
>
> What do you want to do after you have the scores? predict some variables
> from outside those you used to get the scores? find clusters?
>
>
> What constitutes a case?
> How were the cases chosen?
> How much missing data do you have?
> Do you have a unique ID for each case?
> Then  re-post to this list.
>
> shooting from the hip, based on what is done in social indicators work, I
> suggest
> 1) for each battery separately
> do a factor analysis with principal axes and  varimax rotation.
> do a parallel analysis say 10,000 times and overlay the eigenvalues from
> permuted (randomized) data on the scree plot of obtained eigenvalues.
> Pick a few candidate numbers of factors to retain using the factor number
> where the obtained eigen value is at least 1.00 greater than the eigenvalues
> from the randomized data.
> for each candidate number of factors, develop a scoring key using only items
> that load at least .5 or so on the highest factor and not more than .3 or so
> on any other factor.
> Based on reasonableness of interpretation pick a number of factors to
> retain. You will end up dropping some items, some will split across factors,
> and some will simply not load.   You won't necessarily have as many scores
> as originally retained factors. Some factors may not have enough items to
> make a score. Some factors may not make any sense.
> standardize all of the items that go into scores.
> apply the scoring key, i.e. get a mean of all the standardized items that
> went together on a factor add the computed scores to the data file.
> 2) do the same thing on as above but treating the 3 batteries as one set.
>
> Now decide what set of scores/indices makes the most sense in the context
> you intend to use them.
>
> -----------------
>
> if you wanted to find composite scores across all 3 batteries think about In
> the late 60's or early 70 Doug Carroll from Bell Labs published about
> n-battery canonical correlation which found dimensions common  across more
> than the usual 2 batteries for canonical correlation.  This was also a
> generalization of the multitrait multimethod matrix kind of research.
> Although Doug passed away several years ago several years ago, several of
> his students and people developed methods from this approach are on the
> latter 2 discussion lists.
>
> IIRC n-battery canonical correlation was a grandfather to dual scaling,
> correspondence analysis, SEM, etc.
>
> post to
> The Classification Society List
> http://lists.sunysb.edu/index.cgi?A0=CLASS-L
>
> and the Mathematical Psych list
> [hidden email]
>
>
> Art Kendall
> Social Research Consultants
>
> On 7/16/2013 10:55 AM, ro [via SPSSX Discussion] wrote:
>
> Hi Gene,
>
> Thanks for you reply,
>
> I wanna make a index with these results that included this 3 areas. For
> this, I will use the results of score coefficient matrix.
>
> Regards
> Ro
>
> ________________________________
> If you reply to this email, your message will be added to the discussion
> below:
> http://spssx-discussion.1045642.n5.nabble.com/factorial-analysis-results-by-type-of-variables-tp5721185p5721199.html
> To start a new topic under SPSSX Discussion, email [hidden email]
> To unsubscribe from SPSSX Discussion, click here.
> NAML
>
>
> Art Kendall
> Social Research Consultants
>
>
> ________________________________
> If you reply to this email, your message will be added to the discussion
> below:
> http://spssx-discussion.1045642.n5.nabble.com/factorial-analysis-results-by-type-of-variables-tp5721185p5721233.html
> To unsubscribe from factorial analysis results by type of variables, click
> here.
> NAML