choice-metrics.com

Posted: **Tue Mar 13, 2018 4:03 pm**

Hi..
I am facing a difficulty in deciding the attribute levels of some of the attributes considered for alternative specific experiment.

Lets say, Alternative A is likely to have a utility higher than Alternative B (i.e. considering all attributes considered in the design).
Now, there is an attribute x (lets say cost) that is likely to be negatively correlated with the utility of both the alternatives.

There are some other attributes as well, apart from x.

Assumed levels for attribute x are as follows:
(1.) Alternative A: 6, 8 and 10
(2.) Alternative B: 4, 6 and 8

Now, If a few (lets say 2 to 4) of the total 24 choice sets have attribute x's values as, 6 (for Alternative A: the better alternative) and 8 (for Alternative B), it might not look realistic to respondents.

Please advise, if this sort of choice set is okay to be considered? [Note: I have seen a study published in a reputable journal with this sort of choice set present among the total choice sets, however I am wondering if it is a major problem, or it is very common to have one or a few choice sets like that being present].

If this is a major issue, please advise, how to avoid such choice sets from being generated by the design.
Cheers

Posted: **Wed Mar 14, 2018 9:56 am**

Hi

You can always place constraints on a design so that certain attribute combinations do not appear. Whether you impose such constraints will be situation specific however. You state that it may not appear realistic to respondents, but do you know for certain? Having attribute level overlap (where the levels are the same across alternatives) is common in most designs, and in a labelled experiment, may not matter too much, again depending on circumstance. Consider train versus bus. The two alternatives may have the same cost but different travel times. What matters is the betas for cost and time, which may be alternative specific (or generic). You may have theoretical reasons to treat them as generic (your world view aligns more with classical economics where a $ is a $ irrespective of how it is spent), or you may find empirically that people have a different marginal utility for one alternative over the other (this is one benefit of a labelled experiment). If empirically you find the beta for price for bus to be different to train, then the contribution to overall utility will be different assuming the price attribute is the same (i.e., -0.5 * $5 is different to -0.3 *$5). So the question would be is the marginal utility associated with alt A attribute {6,8,10} the same or different to Alt B attribute {4,6,8}. There is no universal answer to this unfortunately - it will be context/study specific.

John

Posted: **Mon Nov 26, 2018 4:31 pm**

Hi

You can always place constraints on a design so that certain attribute combinations do not appear. Whether you impose such constraints will be thesis situation specific however. You state that it may not appear realistic to respondents, but do you know for certain? Having attribute level overlap (where the levels are the same across alternatives) is common in most designs, and in a labelled experiment, may not matter too much, again depending on circumstance. Consider train versus bus. The two alternatives may have the same cost but different travel times. What matters is the betas for cost and time, which may be alternative specific (or generic). You may have theoretical reasons to treat them as generic (your world view aligns more with classical economics where a $ is a $ irrespective of how it is spent), or you may find empirically that people have a different marginal utility for one alternative over the other (this is one benefit of a labelled experiment). If empirically you find the beta for price for bus to be different to train, then the contribution to overall utility will be different assuming the price attribute is the same (i.e., -0.5 * $5 is different to -0.3 *$5). So the question would be is the marginal utility associated with alt A attribute {6,8,10} the same or different to Alt B attribute {4,6,8}. There is no universal answer to this unfortunately - it will be context/study specific.

John

Hello John,

I've been working on the DCE research and seems I've made a mistake in optional features for the task force in alternative specific experiment (excluding generic). A study specific case is related to safety, effectiveness, and mode of administration. As I'm sure about the latter, I can't have a valid confirmation for effectiveness and safety in case of specific alternatives.

Thanks, Renee

choice-metrics.com

Choice set generation using experimental design

Choice set generation using experimental design

Re: Choice set generation using experimental design

Re: Choice set generation using experimental design