Have an idea?

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

Using constant alternatives and avoiding prohibitions

I've been working on a CBC survey for a while and lurking in the forums/reading technical papers. I think I'm getting closer to finishing it off but there's some lingering challenges in the CBC portion that I want to make sure I'm doing correctly. Apologies if these questions are a bit long, I just wanted to provide the proper context.

My survey is a transportation survey asking users to pick between 5 alternatives each time (car, bike, two transit variants, and taxis). I'm primarily varying the transit and taxi alternatives since they're the focus of my research. With this in mind:

1. I base the in-vehicle travel times (IVTT) off their current auto and cycle times (taxi is always the same as auto, transit will be up to 3x longer than auto to account for different stopping patterns). IVTT varies for transit, but not for auto, cycle, or taxi, so I'm mostly using them as reference. For these non-changing attributes, I have two levels for the attribute but nothing changes (I just have a blank space on the second level so that I don't get errors for duplicated level labels, but the output to the user is the exact same following the macOS -- Apple example from the ACBC tutorial).

Is this the right way to go about doing this? And is there still a way to get some idea of how users value that time in comparison to other attributes, or is that something I give up by keeping it static?

2. I have one prohibition currently set up, which is for transfer times and number of transfers on the transit variants. Right now I have it set up to suggest 0-2 transfers, and the times can be from 0-30 minutes, but you can't have 0 transfers and >0 minutes (e.g. 0 transfers and 10 minutes per transfer), because I found my pilot survey testers were rightly getting confused by the combination.

Ideally I'd like to do this without prohibitions to keep it easier to build a more efficient survey instrument, so I was considering collapsing this into one variable (0 transfers, 1 transfer with 0 minutes, 1 transfer with 10, etc.), which would result in somewhere upwards of 9-11 levels based on my current structure. I'm cautious about being able to effectively test these combinations effectively if this becomes 1 attribute with many levels, and about my ability to still test later how sensitive people are to *just* transfer time per transfer or *just* number of transfers.

Is there a recommended way to go about making sure I can still extract this data, or is it better to stick with the prohibition, keeping in mind that I'm also using alternative-specific designs?

Thank you! Much of what I've read here has been immensely useful.
asked Apr 7, 2021 by Jacob (370 points)
retagged Apr 9, 2021 by Jacob

1 Answer

0 votes
Best answer
Had a discussion with Cameron (who was very helpful) about this over the support line, lots of potential options but here's what we discussed and ultimately what I chose to go with:

1. Constant alternatives: I do effectively lose the ability to use in-vehicle time directly when I get to the HB/logit stage because it's constant, but what I can do is either:

a. bin the users together at the end (e.g. people with short/medium/long cycle times and drive times behave in this way), and base their elasticities on just those pools, or:

b. pool all the respondents together to get an aggregate response to IVTT for those attributes.

Both of these would work for me in some way, will be fun to figure it out when I get to that stage!

2. Prohibitions: The solution to this was so clean but I had mild tunnel vision so didn't really think about it this way. Recalling that I have:

Number of transfers: 0, 1, 2, 3
Transfer time: 0 min, 5 min, 10 min, 30 min
Prohibitions: 0 transfers can not appear with 5, 10 or 30 min

Instead of using prohibitions to do this, which messes with my level balance, I can make number of transfers a primary attribute, and set transfer time to only appear with cases of 1, 2, or 3 transfers. When I ran a quick Shortcut design I got perfect level balance, which worked out much better! So now I have two primary attributes: one is the mode (car, cycle, etc.) and one is the number of transfers.
answered Apr 9, 2021 by Jacob (370 points)
edited Apr 9, 2021 by Jacob