Non-parametric multivariate testing

[Jean Hanson]

Hello fellow listers,

I have a dependent variable, HRQOL measure, which is markedly abnormally distributed (and doe not respond to transformation). I would like to look at race, gender and several other factors in a multivariate model. But I do not see a way to accomplish this in SPSS.

 

any ideas?

 

In advance thank you.

 

Jean Hanson

[Peck, Jon]

Without knowing what kind of model you want to estimate, there are two extension commands available that can be useful for linear models with awkward error distributions (note that the distribution of the dependent variable does not need to be normal – it's the error distribution that matters).  The SPSSINC ROBUST REGR and SPSSINC QUANTREG extension commands with dialog box interfaces can be downloaded from SPSS Developer Central (www.spss.com/devcentral).  They require the R and Python plugins and version 17.

 

HTH,

Jon Peck

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Jean Hanson
Sent: Tuesday, May 05, 2009 8:55 AM
To: [hidden email]
Subject: [SPSSX-L] Non-parametric multivariate testing

 

Hello fellow listers,

I have a dependent variable, HRQOL measure, which is markedly abnormally distributed (and doe not respond to transformation). I would like to look at race, gender and several other factors in a multivariate model. But I do not see a way to accomplish this in SPSS.

 

any ideas?

 

In advance thank you.

 

Jean Hanson

[Marta Garcia-Granero]

Jean Hanson wrote:

>
>
> Hello fellow listers,
>
> I have a dependent variable, HRQOL measure, which is markedly
> abnormally distributed (and doe not respond to transformation). I
> would like to look at race, gender and several other factors in a
> multivariate model. But I do not see a way to accomplish this in SPSS.
>
>
>

Conover suggests ranking the data and then running a factorial ANOVA on
the ranks. Graph your data using medians instead of means and you will
get a good idea of the effects of race, gender...

HTH,
Marta García-Granero

--
For miscellaneous SPSS related statistical stuff, visit:
http://gjyp.nl/marta/

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

[Ruben Geert van den Berg]

Dear all,
 
I apologize for this silly question but is there any syntax to display the variable view (like CTRL + T) and value labels instead of values (in variable view)?
 
TIA!
 
Ruben

---

Express yourself instantly with MSN Messenger! MSN Messenger

[Melissa Ives]

Ruben,

 

I think you are looking for the "value labels" icon in the Data Editor toolbar.  It looks like a price tag angled with the point to the top left.

 

It isn't syntax, but then again neither is CTRL + T.

 

If you can't find the icon, go to the View menu and click on Value Labels (the icon is pictured there.

 

Melissa

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ruben van den Berg
Sent: Tuesday, May 05, 2009 1:10 PM
To: [hidden email]
Subject: [SPSSX-L] Syntax to display variable view and value labels instead of values (in variable view)

Dear all,
 
I apologize for this silly question but is there any syntax to display the variable view (like CTRL + T) and value labels instead of values (in variable view)?
 
TIA!
 
Ruben

---

Express yourself instantly with MSN Messenger! MSN Messenger

---

PRIVILEGED AND CONFIDENTIAL INFORMATION
This transmittal and any attachments may contain PRIVILEGED AND
CONFIDENTIAL information and is intended only for the use of the
addressee. If you are not the designated recipient, or an employee
or agent authorized to deliver such transmittals to the designated
recipient, you are hereby notified that any dissemination,
copying or publication of this transmittal is strictly prohibited. If
you have received this transmittal in error, please notify us
immediately by replying to the sender and delete this copy from your
system. You may also call us at [phone] for assistance.

[Peck, Jon]

If you are looking for syntax that can switch between data and variable view or toggle the value labels settings, there isn't any.

 

But you can toggle the value labels on or off using a simple script that could be embedded in a syntax stream using the SCRIPT command.  AFAIK, however, there is no scripting api for switching data and variable view.

 

For the labels, the SpssDataUI class and the GetShowValueLabels method are relevant.  Those are for Python scripting, but similar apis exist for Basic scripting.

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ruben van den Berg
Sent: Tuesday, May 05, 2009 12:10 PM
To: [hidden email]
Subject: [SPSSX-L] Syntax to display variable view and value labels instead of values (in variable view)

 

Dear all,
 
I apologize for this silly question but is there any syntax to display the variable view (like CTRL + T) and value labels instead of values (in variable view)?
 
TIA!
 
Ruben

---

Express yourself instantly with MSN Messenger! MSN Messenger

[William Dudley WNDUDLEY]

Apologies in advance for such a rudimentary question.

I have an ANOVA problem in which I have three binary IV's:
gender,
education (hs vs college)
marital status (married not married)
The Anova command looks like this

UNIANOVA optimism BY  gender mar_status Ed_level
  /METHOD=SSTYPE(3)
  /INTERCEPT=INCLUDE
  /CRITERIA=ALPHA(0.05)
  /DESIGN=gender mar_status Ed_level
         gender*mar_status mar_status*Ed_level gender*ed_level
         gender* mar_status *Ed_level.




Some of the two way interactions are significant  and I want to contrast means within the interactions .
For instance, I have a gender by martial status interaction and I want to compare the means for married and not married separately for males and females.
I could of course just select males and run a t-test using marital status as the IV. BUT I would like to run these tests within the overall ANOVA

Can someone point out how to use the contrast statement (I suppose these a LMATRIX commands) to examine the nature of the interactions at this level?

Thanks in advance for you help.



Bill

[Marta Garcia-Granero]

William Dudley WNDUDLEY wrote:

I have an ANOVA problem in which I have three binary IV's:
gender,
education (hs vs college)
marital status (married not married)
The Anova command looks like this

UNIANOVA optimism BY  gender mar_status Ed_level
  /METHOD=SSTYPE(3)
  /INTERCEPT=INCLUDE
  /CRITERIA=ALPHA(0.05)
  /DESIGN=gender mar_status Ed_level
         gender*mar_status mar_status*Ed_level gender*ed_level
         gender* mar_status *Ed_level.

Some of the two way interactions are significant  and I want to contrast means within the interactions .
For instance, I have a gender by martial status interaction and I want to compare the means for married and not married separately for males and females.

Just kidding... since I'm married, I found very funny your typo error that replaced "marital" by "martial" :-D

I could of course just select males and run a t-test using marital status as the IV. BUT I would like to run these tests within the overall ANOVA

Can someone point out how to use the contrast statement (I suppose these a LMATRIX commands) to examine the nature of the interactions at this level?

I found this short document very useful when I tried LMATRIX time ago:

http://courses.ed.asu.edu/green/Home554/lmatrixA.pdf

Also, here's a copy of part of a message I sent to the list in April 10 (concerning the same question). It is a good idea to search the archives before asking a question to list, sometimes you will find it has already been answered:

--------------------------------
For a 2x2 factorial design, it is quite easy. See this example:

INPUT PROGRAM.
NUMERIC Diet Strain ACholine (F8).
DATA LIST/#N 1 Diet 3 Strain 5.
REPEATING DATA STARTS=6 /OCCURS=#N /DATA ACholine 1-4.
END INPUT PROGRAM.
BEGIN DATA
8 1 1 528 444 338 342 338 331 288 319
8 2 1 434 472 331 444 312 575 575 384
9 1 2 294 175 254 241 352 238 241 269 291
9 2 2 272 388 275 425 350 344 350 425 466
END DATA.
VAR LABEL Strain 'Mouse strain' /ACholine 'AcetylCholine levels'.
VAL LAB Diet 1 'Standard' 2 'Apple'
      /Strain  1 'BalbC'   2 'Alzheimer'.
VAR LEV Diet Strain (NOMINAL).

UNIANOVA ACholine BY Diet Strain
/LMATRIX="Diet effects"
          Diet*Strain -1  0  1  0 Diet -1 1;
          Diet*Strain  0 -1  0  1 Diet -1 1
 /LMATRIX="Strain effect"
         Diet*Strain   1 -1  0  0 Strain 1 -1;
         Diet*Strain   0  0  1 -1 Strain 1 -1
/DESIGN = Diet Strain Diet*Strain .

HTH,
Marta García-Granero
===================== 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]

The question is usually "How are the residuals distributed?". Off hand I cannot think of a multivariate method that requires that the DV itself be "not severely discrepant from normal".

If your residuals are severely discrepant from normal, you may want to use the RANK procedure on the DV and see if it changes your substantive conclusions much. Then see if the methods Jon Peck recommended change your substantive conclusions much.

As an exercise you may want to try all these methods and compare the substantive conclusions, but your reporting will be a lot more straight forward if the residuals from some GLM don't set off an alarm.

Art Kendall
Social Research Consultants

Jean Hanson wrote:

Hello fellow listers,

I have a dependent variable, HRQOL measure,� which is markedly abnormally distributed (and doe not respond to transformation). I would like to look at race, gender and several other factors in a multivariate model. But I do not see a way to accomplish this in SPSS.

�

any ideas?

�

In advance thank you.

�

Jean Hanson