Second choice with some NEW attributes

[Michiel Bliemer]

It is a homogeneous design because it is the same for everyone, correct.
Indeed Ngene currently does not support applying constraints to such designs.

Michiel

[peyman_07]

Many thanks Michiel for your attention.

When we show a set of attributes (real context) and ask respondents to select one alternative and then add some new attributes (hypothetical context) while the previous attribute levels associated with the real setting are fixed and again request them to choose an option amongst the same alternatives, is there any chance of ending up with biased results given that a set of attribute levels (real context) are repeated?

best,
Peyman.

[Michiel Bliemer]

I am not sure why there would be such a bias. You can also choose to change the levels of the existing attributes, but then the comparison has statistically less power so it is a choice you would need to make. If your sample size is large, then you will have sufficient statistical power, but if it is small you will need all the statistical power you can get.

Michiel

[peyman_07]

Hi Michiel,

a) I am not sure about including a price-related attribute in my survey because it might be a dominant attribute. However, it can give me interesting results such as WTP. I intend to minimize the possible risks, so I have come up with an idea. I was wondering if it is possible to include this attribute for half of the sample? Then, if it does not work as expected, we will still have another half to analyze. If it works very well, we will be able to analyze it and also compare the results with the other half?

b) In general, I assume that I need to analyze the collected data separately given that some respondents face an additional attribute. Is that right?

c) Please note that there are two alternatives (binary) and 14 attributes in this design. The sample size is around 500. My concern is about the possibility of ending up with many insignificant parameters due to having a small sample size (250 with that attribute and 250 without it). Would it be the case? If so, please let me know if you have any other suggestion to take advantage of this attribute while minimizing its risks?

Many thanks for your help.
Best,
Peyman.

[Michiel Bliemer]

a) Generally dominant attributes are dealt with through appropriate priors. By having a large negative prior for price, an efficient design would automatically make sure that prices are close together or even constant across all choice tasks for all alternatives. You are proposing to make manual changes. In that case, I would recommend including the price attribute, but setting the levels constant across all alternatives for half of the choice tasks. You can check the section on partial profies in the Ngene manual regarding removing attributes from part of the choice tasks or setting them to the same levels across alternatives for part of the choice tasks.

b) No you can analyse it simultaneously. Check your estimation software how to deal with different numbers of attributes, but generally setting the levels to zero (or another fixed number) across all alternatives means that the attribute drops out (but in case you have interaction terms you need to be careful what you set the level so, I again refer to the partial profile section in the Ngene manual).

c) That could be, it is not possible to say in advance. A pilot study may provide some insights. Are you having 14 attributes or 14 parameters? Having 14 attributes is a lot, can respondents handle that? You may wish to consider partial profile designs.

Michiel

[peyman_07]

Hi Michiel,

Thanks for your response.

a) Based on my understanding of the implicit partial profiles, the analyst is supposed to provide NGENE with a set of omitted attributes in each choice task. It seems to be a type of manual design, right?

b) As mentioned, there are two choices of accepting or declining an offer, so all the attributes including the price are considered as context variables. Therefore, the level of Price is already constant for the alternatives. My concern is that if I include the price, respondents may not consider other attributes. In this design, I guess playing with priors cannot help to address this potential issue, can it?

c) Regarding the number of attributes, I do not show all 14 attributes at the same time. I have divided them into two groups: real and hypothetical situation. In the real context, I show 9 attributes and ask respondents to accept or decline the offer. Then, some additional information is shown to them as hypothetical attributes (this information is not given to respondents in reality) and we see if this info may influence their choice. In this study, Price is a hypothetical attribute. I just want to have two different hypothetical designs: one with the price and the other one without it. I will design both of them in Ngene, so they will be efficient. Then, 50% of the respondents face one design and the other half another one. Using this approach, I will not change the design manually, right? According to this explanation, do you think that this method may not work very well?

Best,
Peyman.

[Michiel Bliemer]

a) The creation of the candidate set is manual, but the generation of an efficient design is not, it uses the modified Federov algorithm to select the most informative choice tasks from the (manually created) candidate set.

b) How you describe it, price is not constant across both alternatives. Accepting an offer means that you pay price $X, while if you do not accept the offer then the price you pay is $0. Not accepting an offer is an opt-out alternative and does not have any attributes. So that would mean that price could indeed be dominant if it is too high. But by not showing the price, what do respondents think that the price is? It is $0? You cannot leave out attributes if they are relevant.

c) It is hard to say without really getting my head around what you are trying to do and whether leaving out the price makes sense (see also my comment above), so I am a bit careful in making recommendations. You may want to sit down with someone who has expertise in designing stated choice experiments and estimating discrete choice models as you seem to have a lot of questions. To make clear recommendations I would need to look at each study in more detail, which is something I cannot do here on the forum.

Michiel