Have an idea?

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

Showing Attribute Labels in Alternative Specific Design... But Not Aligned!

Hi there,

I have an alternative specific CBC about credit cards.  We are testing 3 different types of rates/rewards that are alternative specific.  

1. Percent (e.g., 1% back as reward points)
2. Instant Discount (e.g., 0.5% off instantly)
3. Proprietary Offers (e.g., 5 times normal proprietary rate)

So to do this, I have set up a 3 variable hidden attribute that indicates if it is percent, instant discount, or the proprietary offer and I have alternative specific attributes for each of these 3 types of rewards.  However, I can't show the attribute labels, because each of these three types of rewards are technically different attributes.  

I CAN show the attribute labels if I click "Align Common Attributes on Same Row".  But if I do that, then it shows 3 different rows, all with the same attribute label "Rewards for Using Card" with blank cells where the offer is one alternative rather than the other alternatives (like a tier).  That obviously isn't the right solution, because respondents are supposed to be directly comparing the rewards against each other (e.g., 0.5% instant discount vs. 1% reward points vs. 5 x rate).

The only solution I can think of for this is hide the attribute labels and use html/css to manually build a single column table beside the table with the credit card offers in them.  The only problem with this is that the rows are different heights depending on how much text appears in them, so it will be difficult or impossible to solve the problem this way.

Is there another work around to show the attribute labels this way?
asked Oct 9, 2013 by Joel Anderson Bronze (1,310 points)

1 Answer

0 votes
If each of those conditional attributes has the same number of levels, then you can just them up as a single attribute (rather than three) with "conditional" display.

For example, if each of these 3 alternative-specific attributes has three levels, then you just set it up as a single generic attribute with three different levels, conditional on the credit card type.  Assuming there are also three credit card types, you will display one of 9 pieces of text, depending on the credit card type.  The conditional text replaces the place-holder generic text for that 3-level attribute...all on the same row in the CBC display.

Importantly, during analysis, you specify the interaction between the credit card type and the 3-level attribute.  When you do this, the mathematics are identical between doing it alternative-specific or doing it as a conditional display with an interaction term.
answered Oct 9, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (146,540 points)
Thanks for your feedback, but there are different levels of each ranging from 4 to 6 levels so I can't just combine these with conditional display.

But that's good to know that it's mathematically equivalent to do that.  I will keep that in mind for next time.
One potential solution would be to generate the design as is (separate attributes) and then export that design to a CSV.  Using Excel, recode the attributes down into one large attribute with all of the levels.  Save a copy of your survey, modify it to match the new one large attribute, and import the design.  The display should now be more attractive, but still display the proper concepts to people.  Once you collect your data, you can take the data file from the tweaked design and use the initial survey with the proper design to estimate utilities.
Thanks Brian, that is very nice creative thinking!  I think it's a bit over my head in complexity, partly because I have multiple languages.  But it sounds like theoretically it would work.  Normally I would be happy to try this kind of thing but I am swamped right now and trying to avoid complicating things (although that may be the road I need to go down).

I found another way to do this by combining all 4 variables (type, and the 3 alternative specifica attributes) together as conditional display and only showing "type" with the attribute label from the other 3 (technically I'm tricking the system but it's won't affect the utilities).
I actually thought I have similar problems. I'm making transportation modes analysis, including 4 different modes sharing 3 common attributes which are alternative specific (with 3 levels), such as travel time, travel cost and safety (2 levels).  Could you give an example showing how to "set it up as a single generic attribute with 3 different levels" in this case?

I've tried something like, setting travel costs  in 3  levels (from low to high, but not corresponding to each mode)
1. 2,5,15,35 (I put them on the same row)
2. 3,6,18,40
3. 4,8,20,45

And then in the conditional relationship setting, I displayed travel cost levels corresponding to each mode. Does it make any sense to do this? Or if I'm wrong, please suggest a better solution for dealing with this issue.

Thanks!
Yes, what you are doing is what Bryan suggests (although I didn't understand the context of the 1/2/3 numbers you provided, the written description made sense using the conditional relationships setting).

One caution is that *interpreting* the stand-alone utilities will be difficult or impossible to do, although simulations should be completely fine as per Bryan's mention that it is mathematically equivalent to do this.

If you want to be able to interpret the utilities as a standalone metric (often my clients want to do this) then you can transform the design after fieldwork closes by splitting each of the 3 transportation modes into different attributes (different columns in the design with 0s when it is NA, i.e., for different travel modes that aren't relevant).  Personally, I feel better about the analysis that way, but both should be mathematically equivalent.
Hi Joel,

Thank you so much for your reply. But I just wonder what  the attribute levels could be for each transport mode. For example, I make car as an attribute, so I put car travel time, car travel cost etc. as its levels? Or you meant just setting attributes like: car travel time, PT travel time, car travel cost, PT travel cost...

Sorry I didn't quite figure out how I should proceed.
Car would be the level of another attribute, and you would have a travel time as an attribute and the levels will be generic L1, L2, L3 (if there are 3 levels of travel time).  Then use conditional relationships to re-label them based on the method of transportation (e.g., car).  Hope that helps!
...