Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Re: st: MCA (multiple correspondence analysis), identifying groups


From   jverkuilen <jverkuilen@gc.cuny.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: Re: st: MCA (multiple correspondence analysis), identifying groups
Date   Thu, 14 May 2009 16:31:08 -0400

Classic MCA just involves the singular value decomposition. (Joint CA is a bit trickier because it usually involves an alternating least squares to remove the effect of the main diagonal.) 

So you may be able to trick -biplot- to generate the relevant object scores if you code up the indicator matrix yourself, e.g., by using -xi-. I am not near the manuals to check. 



-----Original Message-----
From: nicola.baldini2@unibo.it
To: statalist@hsphsun2.harvard.edu
Cc: areimondos@gmail.com
Sent: 5/12/2009 12:40 PM
Subject: Re: Re: st: MCA (multiple correspondence analysis),  identifying groups

BTW, something may have changed since my version, but you should be able to download a full version of SPSS (working for 30 days *only*). The problem with -mca- is that it doesn't save transformed data (i.e. the values that each observation takes on your new three dimensions), but some more knowledged with programming should be able to help you to calculate them based on the statements in the ado file. 
Nicola

At 02.33 12/05/2009 -0400, Anna Reimondos wrote:
>Dear Nicola,
>Thank you for your reply and the references! I have downloaded your
>paper and it looks very useful.Unfortunately I do not have SPSS or
>SAS, but I did look for other programs and found the Excel add-in
>'XLSTAT' which does MCA. I am not sure how good it is, but I think it
>was developed by Greenacre so it should be reasonably reliable. As per
>your suggestion I am now trying cluster analysis after the MCA
>(clustering on the coordinates for the two factors that came out of
>the MCA) and so far I the result seem to be making sense.
>I was very confused before as I could not find anything in the
>literature that really discussed what you can do after MCA, but now I
>am much less confused.
>
>Thanks again & have a good day,
>Anna
>
>On Tue, May 12, 2009 at 3:25 AM,  <nicola.baldini2@unibo.it> wrote:
>> I published a paper on which I used MCA (you may download a previous version from SSRN; just search for University Spin-offs and their Environment around there http://ssrn.com/author=410333). First of all, I have to say that I switched to SPSS (and to SAS, for a while) for the analyses. Second, such switch would not solve your problem, since if I remember correctly (my analyses are almost three years old) MCA only builds new variables in lieu of your initial ones, but it doesn't identify groups. A cluster analysis following MCA may identify (which may sound to a reviewer a better grouping device than _you_).
>> You may also read two more empirical papers from which I learnt what is MCA and how to use it:
>> Carayol, N., 2003. Objectives, agreements and matching in science-industry collaborations: Reassembling the pieces of the puzzle. Research Policy 32, 887-908.
>> Carayol, N., Matt, M., 2004. Does research organization influence academic production? Laboratory level evidence from a large European university. Research Policy 33, 1081-1102.
>> A statistical explanation may be found in
>> Greenacre, M., 1993, Correspondence analysis in practice, Academic Press, London.
>> That said, MCA is used to reduce the number of the variables (something, roughly speaking, you may have from PCA and factor analysis, too): if you have four, which reduce to three, is not that gain, since in the meanwhile you loose meaning (i.e. anyone understands what "age" means, but which will be the meaning of the new three dimensions?). If you simply want to group of similar respondents, cluster analysis may be better (discrimant analysis and classification trees may also be useful).
>> Nicola
>>
>> P.S. I'll NOT receive/read any email but the Digest.
>>
>> At 02.33 08/05/2009 -0400, Anna Reimondos wrote:
>>>Hello,
>>>I have a survey dataset consisting of individual respondents and I
>>>would like to group them into different distinct groups.
>>> I have read an article where the researchers had a similar dataset
>>>and they used MCA (multiple correspondence analysis) to identify
>>>distinct groups of respondents. The main variables which distinguish
>>>respondents are their age, sex, marital status and presence of
>>>children.
>>>
>>>I am using STATA 10 , and using the "mca" command I was plugged in my
>>>variables (mca sex age_group marstatus children) and identified 3
>>>dimensions on which I was able to identify 3 clear groupings of people
>>> (for example young unmarried people with no kids, older people with
>>>non-resident kids , and single parents).
>>>Now I would like to use these groupings in further analysis. So for
>>>each respondent I would like a variable that says they are in group
>>>1,2 or 3. I have tried reading the mca help files and looking online
>>>but I cannot understand how to do this. Is there a special command
>>>that is part of the mca command that does this, or do I need to do
>>>this another way.
>>>This is my first time using MCA so I hope this question is not too silly!
>>>
>>>Thanks very much
>>>Anna 

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index