It is often recommend that a 'no choice' option be included in choice-based conjoint, to avoid forcing artificial choices that could result in low quality responses from participants.
Q1. What are the options for analysing this response? You could just omit the 'no-choice' response; however, the refusal to choose contains information that all options in that set have relatively low utilities, compared to sets in a which a choice has been made. How can this information be taken into account when estimating part-worths?
Q2. Are there reasons to include 'no-choice' in best-worst scaling? Has anybody done so and if so how have the above issues been addressed?
I would be grateful for your thoughts.
Hi Malcolm,
1) You can treat the no-choice option just like another alternative. Not sure what software you use for the analysis, but if you use Latent Gold with a 3-file structure system, the only thing you need to do is to accomodate the no-choice option in the "choice" file and the "alternative" file, while for the "set" file the no-choice option will increase your set size by one.
Also, you can always run the analysis with and without the no-choice option and see how much the results differ.
2) Do you mean BW Type 1 or Type 2? In BW Type 1 you don't really need a no-choice option as basically you want a trade off between attributes/features. I am not sure how to handle a BW Type 2 with no-choice (I assume you're saying that the no-choice could be potentially chosen as Best or Worst), and I never did it.
Hope this helps,
Armando
Malcolm,
Malcolm, One can think of best-worse decisions as being fully inclusive of the thought process around the dimension goodness or badness. That is, "this is my worse choice or this is my best choice", the no choice is potentially a measure of another attribute, such as: "I don't find this set/issue relevant/something i can comment on". This seems a different dimension in terms of many studies and might just confound your answers if included in a scale - after all scale questions might/should best be kept to measuring only one dimension.
http://www.rug.nl/research/portal/files/9892552/c4.pdf
discusses this issue in relation to no-choice. The idea that it has low utility information is a bit misleading, as any information is not measured in terms of the utility featured in the "conjoint" test you are making.
Perhaps screen for indifference/relevance/engagement in some other way, if indeed that is your purpose in including no choice? ie how engaged with this category/topic/issue are you? you can still do what Armando suggests regarding splitting responses based on how engaged with the uses in a general sense and then within this scenario ie the nested situation, you can asses their relative choice decisions.
Apologies for not answering your question more directly my bias is to exclude it despite the common practice to include.
Murray
- Badges
- Science topic