SPSSX Discussion

Regression: Multiple DV

Classic

List

Threaded

5 messages Options

Manmit Shrimali-2

Regression: Multiple DV

Hi Team:

This is spss cum stat query. I was wondering if you experts help me
solve the following query:

Background:

I have collected the conjoint data. The design was CVA design i.e.
single concept at a time/per screen. Respondents had to allocate the
chips among 5 drugs summing to 10 chips. We randomly had shown 7 cards
out of 18 cards. Sample size is 196 respondents. We had four attributes
with three levels each.

Query:

Drugs will dependent variables and levels of attributes in conjoint card
will be independent variables. Therefore, we have multiple dependent
variables and multiple independent variables. I want to find out the
impact of independent variable on dependent variable and how IV affects
DV. I cannot take on one DV at a time and regress with IVs as DV will be
having correlations with other DV - increase in prescription for one
drug will obviously reduce the prescription for other drugs which means
pure negative correlation.

Is there any way I can use SPSS to solve the above query? Due to crunch
timeline I will greatly appreciate if you could reply at your earliest.
Using manova will simply tell me any difference exists in groups on DV.
However, I need to get beta coefficients so that I could translate the
impact of each IV on DV. Some suggest that I do regression with one DV
at a time with IV? Any thought on this?

Please advice.

Manmit

paulandpen

Re: Regression: Multiple DV

Manmit,

This is tough (only because I have not done this in spss). In manova (between groups), you can generate beta-weights or parameters by ticking parameter estimates before running the manova. Running separate regressions? Why not the conjoint model in spss since this is how you have generated your design. I am thinking this is a probit or logit type of analysis rather than a traditional regression (linear regression) because the dependent variable is not technically a continuous variable.

There are other people on the list that could share further light on this. Anthony Babinec (sorry to dob you in Anthony) has a way of adjusting the cox regression analysis to potentially maybe deal with your problem. There are some internet links out there on using spss for conjoing analysis.

Best of luck

Paul

> Manmit Shrimali <[hidden email]> wrote:
>
> Hi Team:
>
>
>
> This is spss cum stat query. I was wondering if you experts help me
> solve the following query:
>
>
>
> Background:
>
> I have collected the conjoint data. The design was CVA design i.e.
> single concept at a time/per screen. Respondents had to allocate the
> chips among 5 drugs summing to 10 chips. We randomly had shown 7 cards
> out of 18 cards. Sample size is 196 respondents. We had four attributes
> with three levels each.
>
>
>
> Query:
>
> Drugs will dependent variables and levels of attributes in conjoint card
> will be independent variables. Therefore, we have multiple dependent
> variables and multiple independent variables. I want to find out the
> impact of independent variable on dependent variable and how IV affects
> DV. I cannot take on one DV at a time and regress with IVs as DV will be
> having correlations with other DV - increase in prescription for one
> drug will obviously reduce the prescription for other drugs which means
> pure negative correlation.
>
>
>
> Is there any way I can use SPSS to solve the above query? Due to crunch
> timeline I will greatly appreciate if you could reply at your earliest.
> Using manova will simply tell me any difference exists in groups on DV.
> However, I need to get beta coefficients so that I could translate the
> impact of each IV on DV. Some suggest that I do regression with one DV
> at a time with IV? Any thought on this?
>
>
>
> Please advice.
>
>
>
> Manmit

Manmit Shrimali-2

Re: Regression: Multiple DV

In reply to this post by Manmit Shrimali-2

Paul:

Thanks for your input. I did not generate the design from SPSS as spss
cannot design the CVA allocation based design. Spss will not be able to
analyze the multiple DV. I am also open to using other software but I
need a base platform to begin with :-).

Cheers,
M

-----Original Message-----
From: [hidden email] [mailto:[hidden email]]
Sent: Wednesday, March 21, 2007 8:00 AM
To: Manmit Shrimali
Cc: [hidden email]
Subject: Re: Regression: Multiple DV

Manmit,

This is tough (only because I have not done this in spss). In manova
(between groups), you can generate beta-weights or parameters by ticking
parameter estimates before running the manova. Running separate
regressions? Why not the conjoint model in spss since this is how you
have generated your design. I am thinking this is a probit or logit
type of analysis rather than a traditional regression (linear
regression) because the dependent variable is not technically a
continuous variable.

There are other people on the list that could share further light on
this. Anthony Babinec (sorry to dob you in Anthony) has a way of
adjusting the cox regression analysis to potentially maybe deal with
your problem. There are some internet links out there on using spss for
conjoing analysis.

Best of luck

Paul

attributes
> with three levels each.
>
>
>
> Query:
>
> Drugs will dependent variables and levels of attributes in conjoint
card
> will be independent variables. Therefore, we have multiple dependent
> variables and multiple independent variables. I want to find out the
> impact of independent variable on dependent variable and how IV
affects
> DV. I cannot take on one DV at a time and regress with IVs as DV will
be
> having correlations with other DV - increase in prescription for one
> drug will obviously reduce the prescription for other drugs which
means
> pure negative correlation.
>
>
>
> Is there any way I can use SPSS to solve the above query? Due to
crunch
> timeline I will greatly appreciate if you could reply at your
earliest.
> Using manova will simply tell me any difference exists in groups on
DV.

> However, I need to get beta coefficients so that I could translate the
> impact of each IV on DV. Some suggest that I do regression with one DV
> at a time with IV? Any thought on this?
>
>
>
> Please advice.
>
>
>
> Manmit

Dan Zetu

Re: Regression: Multiple DV

Manmit:

You can run the allocation design using Hierarchical Bayes in Sawtooth
software. If you have used it before, you know that you will need to
generate a chs file which includes the design and allocations. Sawtooth
will output respondent-level utilities which you can use to calculate the
expected share of each drug. Unlike single-choice conjoint, you cannot
directly get the beta coefficients, but you will be able to generate an
Excel-like template to calculate expected shares for each drug.

Do you have access to Sawtooth software?

Dan

>From: Manmit Shrimali <[hidden email]>
>Reply-To: Manmit Shrimali <[hidden email]>
>To: [hidden email]
>Subject: Re: Regression: Multiple DV
>Date: Wed, 21 Mar 2007 08:07:08 +0530
>
>Paul:
>
>Thanks for your input. I did not generate the design from SPSS as spss
>cannot design the CVA allocation based design. Spss will not be able to
>analyze the multiple DV. I am also open to using other software but I
>need a base platform to begin with :-).
>
>Cheers,
>M
>
>-----Original Message-----
>From: [hidden email] [mailto:[hidden email]]
>Sent: Wednesday, March 21, 2007 8:00 AM
>To: Manmit Shrimali
>Cc: [hidden email]
>Subject: Re: Regression: Multiple DV
>
>Manmit,
>
>This is tough (only because I have not done this in spss). In manova
>(between groups), you can generate beta-weights or parameters by ticking
>parameter estimates before running the manova. Running separate
>regressions? Why not the conjoint model in spss since this is how you
>have generated your design. I am thinking this is a probit or logit
>type of analysis rather than a traditional regression (linear
>regression) because the dependent variable is not technically a
>continuous variable.
>
>There are other people on the list that could share further light on
>this. Anthony Babinec (sorry to dob you in Anthony) has a way of
>adjusting the cox regression analysis to potentially maybe deal with
>your problem. There are some internet links out there on using spss for
>conjoing analysis.
>
>Best of luck
>
>Paul
>
>
> > Manmit Shrimali <[hidden email]> wrote:
> >
> > Hi Team:
> >
> >
> >
> > This is spss cum stat query. I was wondering if you experts help me
> > solve the following query:
> >
> >
> >
> > Background:
> >
> > I have collected the conjoint data. The design was CVA design i.e.
> > single concept at a time/per screen. Respondents had to allocate the
> > chips among 5 drugs summing to 10 chips. We randomly had shown 7 cards
> > out of 18 cards. Sample size is 196 respondents. We had four
>attributes
> > with three levels each.
> >
> >
> >
> > Query:
> >
> > Drugs will dependent variables and levels of attributes in conjoint
>card
> > will be independent variables. Therefore, we have multiple dependent
> > variables and multiple independent variables. I want to find out the
> > impact of independent variable on dependent variable and how IV
>affects
> > DV. I cannot take on one DV at a time and regress with IVs as DV will
>be
> > having correlations with other DV - increase in prescription for one
> > drug will obviously reduce the prescription for other drugs which
>means
> > pure negative correlation.
> >
> >
> >
> > Is there any way I can use SPSS to solve the above query? Due to
>crunch
> > timeline I will greatly appreciate if you could reply at your
>earliest.
> > Using manova will simply tell me any difference exists in groups on
>DV.
> > However, I need to get beta coefficients so that I could translate the
> > impact of each IV on DV. Some suggest that I do regression with one DV
> > at a time with IV? Any thought on this?
> >
> >
> >
> > Please advice.
> >
> >
> >
> > Manmit

_________________________________________________________________
Exercise your brain! Try Flexicon.
http://games.msn.com/en/flexicon/default.htm?icid=flexicon_hmemailtaglinemarch07

Manmit Shrimali-2

Re: Regression: Multiple DV

In reply to this post by Manmit Shrimali-2

Dan:

Thanks for your input. The problem is that currently sawtooth does not
offer CVA/HB for allocation type data. It takes only rating for one
dependent variable. I designed the survey in sawtooth and I already
tried that. If this would have been cbc then there r multiple options.
However, options are limited when we are talking about cva constant sum
chip allocation type data.

-----Original Message-----
From: Dan Zetu [mailto:[hidden email]]
Sent: Wednesday, March 21, 2007 8:53 AM
To: Manmit Shrimali; [hidden email]
Subject: Re: Regression: Multiple DV

Manmit:

You can run the allocation design using Hierarchical Bayes in Sawtooth
software. If you have used it before, you know that you will need to
generate a chs file which includes the design and allocations. Sawtooth

will output respondent-level utilities which you can use to calculate
the
expected share of each drug. Unlike single-choice conjoint, you cannot
directly get the beta coefficients, but you will be able to generate an
Excel-like template to calculate expected shares for each drug.

Do you have access to Sawtooth software?

Dan

ticking

>parameter estimates before running the manova. Running separate
>regressions? Why not the conjoint model in spss since this is how you
>have generated your design. I am thinking this is a probit or logit
>type of analysis rather than a traditional regression (linear
>regression) because the dependent variable is not technically a
>continuous variable.
>
>There are other people on the list that could share further light on
>this. Anthony Babinec (sorry to dob you in Anthony) has a way of
>adjusting the cox regression analysis to potentially maybe deal with
>your problem. There are some internet links out there on using spss

for

>conjoing analysis.
>
>Best of luck
>
>Paul
>
>
> > Manmit Shrimali <[hidden email]> wrote:
> >
> > Hi Team:
> >
> >
> >
> > This is spss cum stat query. I was wondering if you experts help me
> > solve the following query:
> >
> >
> >
> > Background:
> >
> > I have collected the conjoint data. The design was CVA design i.e.
> > single concept at a time/per screen. Respondents had to allocate the
> > chips among 5 drugs summing to 10 chips. We randomly had shown 7

cards

> > out of 18 cards. Sample size is 196 respondents. We had four
>attributes
> > with three levels each.
> >
> >
> >
> > Query:
> >
> > Drugs will dependent variables and levels of attributes in conjoint
>card
> > will be independent variables. Therefore, we have multiple dependent
> > variables and multiple independent variables. I want to find out the
> > impact of independent variable on dependent variable and how IV
>affects
> > DV. I cannot take on one DV at a time and regress with IVs as DV

will

>be
> > having correlations with other DV - increase in prescription for one
> > drug will obviously reduce the prescription for other drugs which
>means
> > pure negative correlation.
> >
> >
> >
> > Is there any way I can use SPSS to solve the above query? Due to
>crunch
> > timeline I will greatly appreciate if you could reply at your
>earliest.
> > Using manova will simply tell me any difference exists in groups on
>DV.
> > However, I need to get beta coefficients so that I could translate

the
> > impact of each IV on DV. Some suggest that I do regression with one
DV
> > at a time with IV? Any thought on this?
> >
> >
> >
> > Please advice.
> >
> >
> >
> > Manmit

_________________________________________________________________
Exercise your brain! Try Flexicon.
http://games.msn.com/en/flexicon/default.htm?icid=flexicon_hmemailtaglin
emarch07