Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Independence of attribute levels?

Hi all,

I will really appreciate any advice/suggestions here. I am a PhD student and for my research, I utilized a choice-based conjoint study designed in Lighthouse. I've already ran the experiment and collected the data.

But I'm having some doubts about  my initial design (specifically whether or not there could be attraction effects in attribute levels, and whether or not they are truly independent).

Here is some context for my study. My research aims to investigate the impact of similarity of an individual (based on demographics and/or interests) and group size on decision making in the e-commerce context. To examine this, I used uncertainty-identity theory to propose a research model which I tested using an experimental procedure.

I decided to run the experiment with two separate parts. The first part was a two-way factorial between-subjects design. I adopted a 2 x 4 study to assess the main and interaction effects of two levels of group size and four levels of similarity on decision certainty. The design yielded eight experimental treatments, and fifty-two participants were randomly placed in each of these. The experiment consisted of a task after which respondents were administered a survey. The data was analyzed using partial least squares structural equation modelling.

The second experiment was a choice-based conjoint study/discrete choice experiment. Here, respondents were asked to go through 10 tasks each with 4 concepts and a none option. The attributes were similarity and group size, with 4 levels of similarity and 2 levels of group size (the same levels as in the first part of the experiment). I used a balanced overlap design approach.

The levels were as follows:

(1) No similarity
(2) Similarity in demographics
(3) Similarity in interests
(4) Similarity in both demographics and interests

Group size:
(1) Small group size (10 people in the group)
(2) Large group size (150 people in the group)

I am concerned that perhaps the attribute levels for similarity are not mutually exclusive, because one of the levels is basically a combination of two other levels. And could there be attraction effects for the levels in group size? If this is the case, would be my data be unusable for the conjoint portion?

Here is a link to the study: https://resortbooker.sawtoothsoftware.com/login.html

Thank you for your help with this.
asked Mar 10 by munawarm (120 points)
edited Mar 11 by munawarm
Sometimes people with different academic backgrounds use different terms in conjoint analysis.  Could you clarify what you mean by "could there be attraction effects" in your question?
Hi Brian, thank you for responding. Attraction effects is something I read in Bryan Orme's book "Getting Started with Conjoint Analysis" where he defines it as the "psychological context effect that causes an inferior product to increase the attractiveness of a superior one" (pp. 169).  I guess I was wondering if having two varying levels of group size (one obviously bigger than the other) could potentially affect utility estimates.

My other larger concern is about whether or not my attribute levels for similarity can be considered independent. One of the levels is basically a combination of two other levels. Could this be a problem? I don't want to have to recollect all my data, and I'm just hoping there is a way to work around this.

Thanks again for taking the time to read through this.
It's ok to have one level that is a combination of others.  When we say that they are independent, we mean more like they can vary independently of each other.  For example, it would be odd to have one level be 1-2 year warranty and another level to be 2-3 year warranty.

The attraction effects are I think documented pretty well in behavior economics, though in conjoint analysis we hope this isn't a big concern since we try to show an equal number of combinations of attribute levels so ideally there isn't a series of objectively good versus objectively bad options on each page.

So, I'm not sure if I can offer anything concrete to you.  I don't think what you've done is wrong or anything, but if I were giving advice I would probably mentions that it can be difficult to use generic terms in conjoint analysis, like "similarity in demographics" might mean different things to different people.  Are those demographics age, race, religion, income, etc.?  What exactly is in the respondent's mind when they are evaluating this?  It's more difficult than concrete things like "red" or "$20" or "1 year warranty."
Hi Brian thanks for your response. For me the biggest concern was having a level that was a combination of two others- so I'm happy that's not going to be a problem.

 In terms of the actual way I wrote out the levels- I think it's clear (but then again, I'm biased!). I don't know if this link works, but if it does, please have a look if you get a chance, and let me know what you think!


Thanks for all your help here!
Sure, like I don't think the task is particularly confusing or it's not possible for people to give you a good answer, just that conjoint models are better predictors of real behavior the more specific/concrete the levels are.  This could also probably improve the none, which is a bit ambiguous (does choosing none mean "I would not go on this trip?").

Your solution to the original question

Please only use this to answer the original question. Otherwise please use comments.
Your name to display (optional):
Privacy: Your email address will only be used for sending these notifications.
Anti-spam verification:

To avoid this verification in future, please log in or register.