From Lixian.Qian at xjtlu.edu.cn  Tue Feb 16 12:57:48 2021
From: Lixian.Qian at xjtlu.edu.cn (Lixian Qian)
Date: Tue, 16 Feb 2021 01:57:48 +0000
Subject: [Limdep Nlogit List] About dominance in D-optimal design
Message-ID: <e39ac47f5d584b49a506d832c2f65055@xjtlu.edu.cn>

Hello, DCM experts,

I am not sure whether this is the best place to ask a question about alternative dominance in DCM experiment design, but I hope you can give me some advices.

We have a D-optimal stated preference experiment design with 4 alternatives. We find in some scenarios, alternative B dominates alternative A.
We are not sure whether the scenarios with alternative dominance should be removed from the design. If removed, will the design still be optimal?

We had some pilot data based on the DCM experiment design. No matter whether we remove scenarios with alternative dominance, the choice model analysis based on NLogit does not show significant differences on key attributes. Does it mean it is acceptable for us to keep all scenarios in the formal data collection?

Thank you.


Best Regards,
Lixian


From teagle at tceagle.com  Wed Feb 17 04:06:40 2021
From: teagle at tceagle.com (Thomas Eagle)
Date: Tue, 16 Feb 2021 17:06:40 +0000
Subject: [Limdep Nlogit List] About dominance in D-optimal design
In-Reply-To: <e39ac47f5d584b49a506d832c2f65055@xjtlu.edu.cn>
References: <e39ac47f5d584b49a506d832c2f65055@xjtlu.edu.cn>
Message-ID: <d7365658b7624a6ab4ac65c99203462e@tceagle.com>

Lixian,

Technically the dominated alternative is only required to make the design more d-optimal.  If you are willing to sacrifice some d-optimality you can remove it.  Having a dominant alternative in a task is not useful information if the alternative it dominates is never chosen and as such it is a waste of task/design space.  The dominated and dominant alternative adds no useful information to the model estimation.  I typically build the design such that domination does not occur, but that sacrifices d-optimality.  A short cut approach is to modify by hand the dominate alternative so that is does not dominate any longer.  I would only do that if you have no means to restrict the initial or final design to exclude such dominance.

Tom Eagle

-----Original Message-----
From: Limdep <limdep-bounces at mailman.sydney.edu.au> On Behalf Of Lixian Qian
Sent: Monday, February 15, 2021 5:58 PM
To: limdep at limdep.itls.usyd.edu.au
Subject: [Limdep Nlogit List] About dominance in D-optimal design

Hello, DCM experts,

I am not sure whether this is the best place to ask a question about alternative dominance in DCM experiment design, but I hope you can give me some advices.

We have a D-optimal stated preference experiment design with 4 alternatives. We find in some scenarios, alternative B dominates alternative A.
We are not sure whether the scenarios with alternative dominance should be removed from the design. If removed, will the design still be optimal?

We had some pilot data based on the DCM experiment design. No matter whether we remove scenarios with alternative dominance, the choice model analysis based on NLogit does not show significant differences on key attributes. Does it mean it is acceptable for us to keep all scenarios in the formal data collection?

Thank you.


Best Regards,
Lixian

_______________________________________________
Limdep site list
Limdep at mailman.sydney.edu.au
https://protect-au.mimecast.com/s/doELCr81nytk8yxgU7SN2q?domain=limdep.itls.usyd.edu.au


From teagle at tceagle.com  Wed Feb 17 04:06:40 2021
From: teagle at tceagle.com (Thomas Eagle)
Date: Tue, 16 Feb 2021 17:06:40 +0000
Subject: [Limdep Nlogit List] About dominance in D-optimal design
In-Reply-To: <e39ac47f5d584b49a506d832c2f65055@xjtlu.edu.cn>
References: <e39ac47f5d584b49a506d832c2f65055@xjtlu.edu.cn>
Message-ID: <d7365658b7624a6ab4ac65c99203462e@tceagle.com>

Lixian,

Technically the dominated alternative is only required to make the design more d-optimal.  If you are willing to sacrifice some d-optimality you can remove it.  Having a dominant alternative in a task is not useful information if the alternative it dominates is never chosen and as such it is a waste of task/design space.  The dominated and dominant alternative adds no useful information to the model estimation.  I typically build the design such that domination does not occur, but that sacrifices d-optimality.  A short cut approach is to modify by hand the dominate alternative so that is does not dominate any longer.  I would only do that if you have no means to restrict the initial or final design to exclude such dominance.

Tom Eagle

-----Original Message-----
From: Limdep <limdep-bounces at mailman.sydney.edu.au> On Behalf Of Lixian Qian
Sent: Monday, February 15, 2021 5:58 PM
To: limdep at limdep.itls.usyd.edu.au
Subject: [Limdep Nlogit List] About dominance in D-optimal design

Hello, DCM experts,

I am not sure whether this is the best place to ask a question about alternative dominance in DCM experiment design, but I hope you can give me some advices.

We have a D-optimal stated preference experiment design with 4 alternatives. We find in some scenarios, alternative B dominates alternative A.
We are not sure whether the scenarios with alternative dominance should be removed from the design. If removed, will the design still be optimal?

We had some pilot data based on the DCM experiment design. No matter whether we remove scenarios with alternative dominance, the choice model analysis based on NLogit does not show significant differences on key attributes. Does it mean it is acceptable for us to keep all scenarios in the formal data collection?

Thank you.


Best Regards,
Lixian

_______________________________________________
Limdep site list
Limdep at mailman.sydney.edu.au
https://protect-au.mimecast.com/s/8LhnCwV1vMfqG5owUVeC0I?domain=limdep.itls.usyd.edu.au


From doctorjasonong at gmail.com  Tue Feb 23 14:28:58 2021
From: doctorjasonong at gmail.com (Jason Ong)
Date: Tue, 23 Feb 2021 14:28:58 +1100
Subject: [Limdep Nlogit List] calculating standard deviations for reference
 level in RPL model?
Message-ID: <CAG6D8PD0vEoX2MhUam92=4B21_C0vR8fixrz28SXKar39hXGFw@mail.gmail.com>

hi,

Is it possible to output/calculate the standard deviation from the
reference level of an attribute?
For example, part of my RPL model is shown below (where I used effects
coding) but because the NLOGIT output doesn't report the standard deviation
(or standard error of the standard deviation), how can I calculate that?

Attribute

Coefficient



Standard deviation



Cost









-       No cost (reference)

0.75

***





-       500 Naira

0.16

***

0.23

**

-       1000 Naira

-0.21

***

0.36

***

-       2000 Naira

-0.70

***

0.55

***


*Jason Ong*
Twitter: @DrJasonJOng
PhD, MMed, MBBS, FAChSHM, FRACGP

Sexual Health Physician, Melbourne Sexual Health Centre (MSHC)
Head, HIV/STI Economics and Health Preference Research (MSHC)

Associate Professor (Hon), London School of Hygiene and Tropical Medicine,
UK
Central Clinical School, Monash University, Australia
Melbourne School of Population and Global Health, University of Melbourne,
Australia

Associate Editor, Sexually Transmitted Infections
Special Issues Editor, Sexual Health
Board Director, ASHM (https://protect-au.mimecast.com/s/1tw8C81V0PT2APD1sn1LRG?domain=ashm.org.au)
Board Director, AFAO (https://protect-au.mimecast.com/s/GApiC91WPRTl6NLQhEn1UD?domain=afao.org.au)
My publications: https://protect-au.mimecast.com/s/qliGC0YKPvi79gqVC2CsSy?domain=lshtm.ac.uk