Questions about the conditions of the CE respondents

This forum is for posts covering broader stated choice experimental design issues.

Moderators: Andrew Collins, Michiel Bliemer, johnr

Questions about the conditions of the CE respondents

Postby Lohas » Thu Oct 10, 2019 2:36 pm

Dear Michiel,
I have some questions about the condition and the requirement for the respondents.

As my research is going to test the Chinese consumer preferences toward different e-commerce channels (supply chains) of Australian imported beef, I have some Qs about the condition of the respondents:
1. Does the respondent have to be an Australian imported beef consumer/ imported beef consumer/ beef consumer/ meat consumer/ any consumer?
2. As the aim of the research is to test the consumer preferences of buying behavior via different e-commerce channels and omni-channels, does the consumer have to be someone who had purchased beef online before?

Many thanks in advance.
Kind regards,
Lohas
Lohas
 
Posts: 17
Joined: Fri Sep 20, 2019 8:36 pm

Re: Questions about the conditions of the CE respondents

Postby Michiel Bliemer » Thu Oct 10, 2019 3:34 pm

1. That depends on your research question. But I think that asking vegetarians about beef consumption does not make sense.
2. No, in a stated choice experiment you can task for new alternatives that may not yet exist or have been considered. For example, if the price of purchasing beef online is much cheaper than purchasing beef in a regular store, then you are asking people to make a trade-off on cost and purchase channel.

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm

Re: Questions about the conditions of the CE respondents

Postby Lohas » Thu Oct 10, 2019 4:16 pm

Michiel Bliemer wrote:1. That depends on your research question. But I think that asking vegetarians about beef consumption does not make sense.
2. No, in a stated choice experiment you can task for new alternatives that may not yet exist or have been considered. For example, if the price of purchasing beef online is much cheaper than purchasing beef in a regular store, then you are asking people to make a trade-off on cost and purchase channel.

Michiel


Dear Michiel,
Really appreciate for your reply!!! That is really helpful and has addressed a lot of my concerns.
Another questions are:
1. My research questions are "What the Chinese consumers' preferences of buying Australian imported beef in an e-commerce and omni-channel environment?", "What's their WTP for the different channel levels, traceability attribute levels, and Country of origin attribute levels?"
Therefore, regarding your above answers, the respondent can also be consumers who don't have the purchase experience of Australian imported beef, or any other countries imported beef, or Australian imported beef that selling online, right?
2. I am going to conduct the main survey in 4 different 1st tier cities in China. But, if I use an D-efficient design (the D-efficient design is efficient design, right>) as you suggested before, do I need to get the prior 4 times in 4 different cities? or I can get the pilot study done in just 1 city, if yes, can that 1 city be a 5th city which is not chosen from those 4 cities (4 main survey cities)? What if the 5th city chosen for pilot study is still one city in China, but may get different economic scales, dietary habits, and etc.?

Best regards,
Lohas
Lohas
 
Posts: 17
Joined: Fri Sep 20, 2019 8:36 pm

Re: Questions about the conditions of the CE respondents

Postby Michiel Bliemer » Thu Oct 10, 2019 4:49 pm

1. I think that is fine.
2. If you think that behaviour in different cities is more or less the same, then you can just do a pilot study in a single city and use those priors for all cities. Preferably your pilot study covers respondents from all cities.

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm

Re: Questions about the conditions of the CE respondents

Postby Lohas » Thu Oct 10, 2019 5:18 pm

Michiel Bliemer wrote:1. I think that is fine.
2. If you think that behaviour in different cities is more or less the same, then you can just do a pilot study in a single city and use those priors for all cities. Preferably your pilot study covers respondents from all cities.

Michiel


Dear Michiel,

Many thanks!

1. So, to conclude, the respondent doesn't have to be someone who had purchased beef before; also the respondent can be someone who doesn't have online shopping experience before.
2. In term of the "Price" attribute, as I see from our training textbook, the interval between prices should be neither too big, nor too small. Otherwise, it will put risks to the results, and come up with a parameter for price that turn out not significant? So, how to set the intervals of "price attribute", so that it won't come up with a parameter which is not significant at the end?
3. What if the result come up with a not significant parameter for Price attribute? Will the WTP still work in this situation?
4. In terms of the different analysis models used for the data analysis, such as, MNL, RPL (mixed logit model), and etc, when designing the questionnaire in Ngene, the syntax should be all the same, no matter which model I chose for the data analysis later, right? The distinction process of which model to be used for analysis, is only lying in the Nlogit analysis process, right?

Many thanks in advance and looking forward to your reply.

Kind regards,
Lohas
Lohas
 
Posts: 17
Joined: Fri Sep 20, 2019 8:36 pm

Re: Questions about the conditions of the CE respondents

Postby Michiel Bliemer » Fri Oct 11, 2019 8:23 am

1. Yes
2. The range should be as wide as possible while keeping it realistic.
3. Then your WTP measure will also not be statistically significant (i.e., have a very large standard error).
4. It is difficult to design data specifically for models other than MNL, so it is common to design the data for estimating the MNL model but you can test your design for estimating the RPL model. You can test for this in Ngene by specifying multiple models and only optimise on the MNL model but allow evaluation for other models, e.g.

;eff = mnl_model(mnl,d)
;model(mnl_model):
...
;model(rppanel_model):
...

Of course you will need priors for the RPL model, which are often difficult to obtain from a pilot study.

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm

Re: Questions about the conditions of the CE respondents

Postby Lohas » Sat Oct 12, 2019 2:54 pm

Michiel Bliemer wrote:1. Yes
2. The range should be as wide as possible while keeping it realistic.
3. Then your WTP measure will also not be statistically significant (i.e., have a very large standard error).
4. It is difficult to design data specifically for models other than MNL, so it is common to design the data for estimating the MNL model but you can test your design for estimating the RPL model. You can test for this in Ngene by specifying multiple models and only optimise on the MNL model but allow evaluation for other models, e.g.

;eff = mnl_model(mnl,d)
;model(mnl_model):
...
;model(rppanel_model):
...

Of course you will need priors for the RPL model, which are often difficult to obtain from a pilot study.

Michiel


Many thanks Michiel,
Regarding your reply,

1. Regarding previous Q2: What if the consumers only make trade offs between price, and ignore other attribute, if the range between prices is too large? For example, I set price attribute as $10, $20, $30, and $40. Then, come up with a choice task that with Alternative 1: $10, comparing with Alternative 2 which is $40. In this case, I guess most the common consumers may choose Alternative 1, because of cheap and ignore other attributes.
2. Regarding previous Q4: So that means, no matter which model or models I am going to use for my data analysis, the Ngene design are exactly the same, right?
For example, I planed to use the RPL model for my study when I am designing the questionnaire, however, after I collected the data, I come up with a idea that to use both MNL, RPL and ELC, or other models together. So, in this situation, the same design (Ngene syntax) for DCE questionnaire in Ngene and the same data which have been collected, can be used for analyzing different models, right?

Many thanks Michiel,
Best regards,
Lohas
Lohas
 
Posts: 17
Joined: Fri Sep 20, 2019 8:36 pm

Re: Questions about the conditions of the CE respondents

Postby Michiel Bliemer » Mon Oct 14, 2019 9:02 am

1. Yes that could happen with inappropriate or non-informative (zero) priors. If your priors reflect the fact that price is a dominant attribute then this situation will automatically be avoided with appropriate priors since a choice task where one alternative is chosen by, say, 99% of the people is not very efficient and therefore automatically discarded in the design generation process. But good priors are key.

2. Yes, one typically generates a design that is efficient for estimating an MNL model and then later estimates more advanced models on the same dataset, assuming that it is also reasonably efficient for more advanced models.

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm

Re: Questions about the conditions of the CE respondents

Postby Lohas » Mon Oct 14, 2019 11:24 am

Michiel Bliemer wrote:1. Yes that could happen with inappropriate or non-informative (zero) priors. If your priors reflect the fact that price is a dominant attribute then this situation will automatically be avoided with appropriate priors since a choice task where one alternative is chosen by, say, 99% of the people is not very efficient and therefore automatically discarded in the design generation process. But good priors are key.

2. Yes, one typically generates a design that is efficient for estimating an MNL model and then later estimates more advanced models on the same dataset, assuming that it is also reasonably efficient for more advanced models.

Michiel


Many thanks Michiel,

1. As I am going to use the "orth=ood" design in my study, which means no priors, so how to make sure that the price attribute can work?
2. What is the logics of choosing different models for analyse? I see many paper say if you want to test the heterogeneity of consumers, you need to use RPL model (in my research domain, many researcher using this RPL in their similar studies), but I see some papers using 2 models such as RPL + ECM. So how to decide which model is fit for your study?

Kind regards,
Lohas
Lohas
 
Posts: 17
Joined: Fri Sep 20, 2019 8:36 pm

Re: Questions about the conditions of the CE respondents

Postby Michiel Bliemer » Mon Oct 14, 2019 11:44 am

1. It will work but there may be several weakly dominant alternatives because of your price attribute. This cannot be avoided with an orthogonal design.

2. That is a model estimation question, not an experimental design question. Most people estimate first an MNL, and then an RPL panel or latent class panel model to account for unobserved heterogeneity and look at the adjusted rho squared and the parameter estimates to see which model they like best. You may want to post that question on a forum of model estimation software (Nlogit, Biogeme, Apollo, ...).

Michiel
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm

Next

Return to Choice experiments - general

Who is online

Users browsing this forum: No registered users and 3 guests

cron