MULTIPLE correspondence analysis with four MULTI response sets: how does this work?

[Mat Hematic]

Dear SPSS-users,

unfortunately, I am not able to run a MULTIPLE correspondence analysis with four MULTI response sets. In the variable list of the SPSS – submenu „optimal scaling“, multiple response sets are not shown.

Could it be, that it is not possible, to conduct a MULTIPLE correspondence analysis with MULTI response sets directly or through data transformation? Binary "yes"/"no"-variables are no solution, because: the "no"-categories should not influence and should not be included in the joint space, and because there should not be a dimensionality of aroung 50 (-> 50 categories of 4 multiple response sets), but 4, as the number oft multiple response sets used.

Many thanks for your help.

Best regards
Mat

[PRogman]

Did you define your Multiple Response Set (Main menu: Analyze|Multiple Response|Define Variable Sets...) before trying to run the procedure? Please see TFM on Multiple Repsonse Analysis.

FILE HANDLE fpSPSSSamples /NAME='C:\Program Files\SPSS\22\Samples\English'.
GET FILE='fpSPSSSamples \survey_sample.sav'.
DATASET NAME Survey_Sample WINDOW=FRONT.
MULT RESPONSE
  GROUPS=$News 'News Source' (news1 news2 news3 news4 news5 (1))
  /FREQUENCIES=$News.

HTH, PRogman

Mat Hematic wrote

Dear SPSS-users,

unfortunately, I am not able to run a MULTIPLE correspondence analysis with four MULTI response sets. In the variable list of the SPSS – submenu „optimal scaling“, multiple response sets are not shown.

Could it be, that it is not possible, to conduct a MULTIPLE correspondence analysis with MULTI response sets directly or through data transformation? Binary "yes"/"no"-variables are no solution, because: the "no"-categories should not influence and should not be included in the joint space, and because there should not be a dimensionality of aroung 50 (-> 50 categories of 4 multiple response sets), but 4, as the number oft multiple response sets used.

Many thanks for your help.

Best regards
Mat

[Mat Hematic]

Dear PRogman,

thank you for your message! Yes, the multi response set is already existing, but as written, it seems not possible to run with multi response sets a multiple correspondence analysis.

Best regards
Mat

[David Marso]

This might be useful?
http://stats.stackexchange.com/questions/68938/multiple-correspondence-analysis-for-count-data-entered-as-binary-variables                       

Mat Hematic wrote

Dear SPSS-users,

unfortunately, I am not able to run a MULTIPLE correspondence analysis with four MULTI response sets. In the variable list of the SPSS – submenu „optimal scaling“, multiple response sets are not shown.

Could it be, that it is not possible, to conduct a MULTIPLE correspondence analysis with MULTI response sets directly or through data transformation? Binary "yes"/"no"-variables are no solution, because: the "no"-categories should not influence and should not be included in the joint space, and because there should not be a dimensionality of aroung 50 (-> 50 categories of 4 multiple response sets), but 4, as the number oft multiple response sets used.

Many thanks for your help.

Best regards
Mat

[Mat Hematic]

Thank you for your efforts. I am working with 4 multiple response sets (not just two as mentioned in the link), e.g. brand x attributes x costumer needs x information behavior. Thus, I really have to run a multiple correspondence analysis.

Many thanks for all your help in advance!

Best regards
Mat

[David Marso]

So generalize the concept and carry on!

Mat Hematic wrote

Thank you for your efforts. I am working with 4 multiple response sets (not just two as mentioned in the link), e.g. brand x attributes x costumer needs x information behavior. Thus, I really have to run a multiple correspondence analysis.

Many thanks for all your help in advance!

Best regards
Mat

[Mat Hematic]

All 3 methods to run a corresponse analyses seem not to work with more than 2 multi response sets:

1) the "syntax"-format is based on a 2-dimensional contingency table (rows and columns). A third and fourth dimension can not be displayed / integrated in a syntax file

2) the "weight"-format is based on a 2-dimensional contingency table (rows and columns) and is tranformed into 3 variables: Variable "columns", variable "rows" and variable "weight = frequencies". e.g.:
                  Brand1   Brand2    Brand3
Attribute1        5           4            2
Attribute2        7           1            6
Attribute3        9           3            8
...

-> transformation into 3 variables, in order to run the correspondence analysis ("weight"-variable has to be activated before running the correspondence analyis):

column-Var           row-Var       weight (=frequency)
      1                        1                           5
      1                        2                           7
      1                        3                           9
      2                        1                           4
      2                        2                           1
      2                        3                           3
      3                        1                           2
      3                        2                           6
      3                        3                           8

It seems not possible to do such a transformation based on 3 or 4 dimensions (if each of the 4 multiple response sets  has only 10 categories, you would get 10x10x10x10=10'000 combinations mostly with weights = 0 (instead of 9 combinations in the example above).

3) also the third way to run a correspondence analysis is not possible. multiple response sets are not shown in the variable list of the (multiple) corresponse analysis.

[David Marso]

The following code 'explodes' the coincidence patterns.
HTH
SET MPRINT ON PRINTBACK ON.
DATA LIST LIST / ID MD1_1 MD1_2 MD1_3 MD2_1 MD2_2 MD2_3 MD3_1 MD3_2 MD3_3 .
BEGIN DATA
1  1 0 1   1 0 1   0 1 1
2  0 1 1   1 1 0   1 1 0
3  0 1 1   1 1 0   0 1 1
4  1 1 1   0 1 0   1 1 1
END DATA.
VECTOR
   D1=MD1_1 TO MD1_3
 / D2=MD2_1 TO MD2_3
 / D3=MD3_1 TO MD3_3.
LOOP I1=1 TO 3.
DO IF D1(I1)=1.
LOOP I2=1 TO 3.
DO IF D2(I2)=1.
LOOP I3=1 TO 3.
DO IF D3(I3)= 1.
XSAVE OUTFILE "C:\TEMP\scratch.sav" / KEEP ID I1 I2 I3.
END IF.
END LOOP.
END IF.
END LOOP.
END IF.
END LOOP.
EXECUTE.
GET FILE "C:\TEMP\scratch.sav" .
LIST.


      ID       I1       I2       I3
 
    1.00     1.00     1.00     2.00
    1.00     1.00     1.00     3.00
    1.00     1.00     3.00     2.00
    1.00     1.00     3.00     3.00
    1.00     3.00     1.00     2.00
    1.00     3.00     1.00     3.00
    1.00     3.00     3.00     2.00
    1.00     3.00     3.00     3.00
    2.00     2.00     1.00     1.00
    2.00     2.00     1.00     2.00
    2.00     2.00     2.00     1.00
    2.00     2.00     2.00     2.00
    2.00     3.00     1.00     1.00
    2.00     3.00     1.00     2.00
    2.00     3.00     2.00     1.00
    2.00     3.00     2.00     2.00
    3.00     2.00     1.00     2.00
    3.00     2.00     1.00     3.00
    3.00     2.00     2.00     2.00
    3.00     2.00     2.00     3.00
    3.00     3.00     1.00     2.00
    3.00     3.00     1.00     3.00
    3.00     3.00     2.00     2.00
    3.00     3.00     2.00     3.00
    4.00     1.00     2.00     1.00
    4.00     1.00     2.00     2.00
    4.00     1.00     2.00     3.00
    4.00     2.00     2.00     1.00
    4.00     2.00     2.00     2.00
    4.00     2.00     2.00     3.00
    4.00     3.00     2.00     1.00
    4.00     3.00     2.00     2.00
    4.00     3.00     2.00     3.00
 
 
Number of cases read:  33    Number of cases listed:  33

[Mat Hematic]

Hello! Thank you, but unfortunately I am not understanding how to solve the problem with that syntax. Is the syntax really a practical way for 20 x 20 x 20 x 20 multiple response sets (= thousands of category combinations)? Is there no other way existing? Thus, help is still needed

Thanks again and best regards
Mat

[David Marso]

What is there to not understand?
It is a few simple loops with a file write in the inner loop.
YES!  This is probably the ONLY practical way to barf out all those combinations.

ALL YOU NEED TO DO IS ADD AN ADDITIONAL LOOP AND DO IF and change
the dimensions to 20 rather than 3.
VECTOR, LOOP and DO IF are very well documented in the Fine Manual.
Assuming I am on the right track and to apply MCA you will need the patterns
of juxtaposition of each set?
Other than that maybe you need to be clearer on how you need the data represented.
Of course as soon as one ventures into multiple response sets all sorts of problems arise for doing any sort of meaningful statistical analysis.
How do others frame the MCA problem within the context of multiple response sets?
How do they deal with the fact that the responses are NOT independent?
Isn't Correspondence analysis essentially a decomposition of a Chi Square?
Doesn't such an analysis assume the rows and columns are INDEPENDENT?
See isn't it all so clear and simple once you throw away important assumptions?

---

Mat Hematic wrote

Hello! Thank you, but unfortunately I am not understanding how to solve the problem with that syntax. Is the syntax really a practical way for 20 x 20 x 20 x 20 multiple response sets (= thousands of category combinations)? Is there no other way existing? Thus, help is still needed

Thanks again and best regards
Mat

[Mat Hematic]

Dear David,

many thanks for your support. To be honest, I am really not that strong with SPSS syntax. Howewer: I am dividing the MCA into 2 or 3 correspondence analysis. This has the advantage, that the axes can be interpreted easier.

Thank you again, best regards
Mat



What is there to not understand?
It is a few simple loops with a file write in the inner loop.
YES!  This is probably the ONLY practical way to barf out all those combinations.

ALL YOU NEED TO DO IS ADD AN ADDITIONAL LOOP AND DO IF and change
the dimensions to 20 rather than 3.
VECTOR, LOOP and DO IF are very well documented in the Fine Manual.
Assuming I am on the right track and to apply MCA you will need the patterns
of juxtaposition of each set?
Other than that maybe you need to be clearer on how you need the data represented.
Of course as soon as one ventures into multiple response sets all sorts of problems arise for doing any sort of meaningful statistical analysis.
How do others frame the MCA problem within the context of multiple response sets?
How do they deal with the fact that the responses are NOT independent?
Isn't Correspondence analysis essentially a decomposition of a Chi Square?
Doesn't such an analysis assume the rows and columns are INDEPENDENT?
See isn't it all so clear and simple once you throw away important assumptions?

[David Marso]

VECTOR
   D1=MD1_1 TO MD1_20
 / D2=MD2_1 TO MD2_20
 / D3=MD3_1 TO MD3_20
 / D4=MD4_1 TO MD4_20.
LOOP I1=1 TO 20.
DO IF D1(I1)=1.
LOOP I2=1 TO 20.
DO IF D2(I2)=1.
LOOP I3=1 TO 20.
DO IF D3(I3)= 1.
LOOP I4=1 TO 20.
DO IF D4(I4)= 1.

XSAVE OUTFILE "C:\TEMP\scratch.sav" / KEEP ID I1 I2 I3 I4.
END IF.
END LOOP.
END IF.
END LOOP.
END IF.
END LOOP.
END IF.
END LOOP.
EXECUTE.
GET FILE "C:\TEMP\scratch.sav" .
LIST.