Have an idea?

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

Can two conditional attributes from different alternative specific constants interact?

"Conditional attribute: attributes only displayed with a particular level or levels of the primary attribute."

"You can specify two-way interactions for alternative-specific designs during analysis, though in our example it doesn't make sense to specify two-way interactions between primary and conditional attributes, since the effects measured are already "specific" to the reference levels of the primary attribute.  However, for this example design, it is possible to study the interaction between the frequency of bus pick-up and the cost per bus trip, since these two attributes always were displayed together within concepts (and no prohibitions involving these two attributes were specified).  If a common attribute were defined (applying to all levels of the primary attribute), it is possible to specify other interactions, such as between the primary and common attribute."

https://www.sawtoothsoftware.com/help/lighthouse-studio/manual/hid_web_cbc_designs_7.html

But what about interactions between conditional attributes that occur within different alternatives? Are those possible to detect in lighthouse?
asked Apr 20, 2020 by Kelly

1 Answer

0 votes
So for a transportation example, you might have car, bus, drive.  Bus has a "wait time" attribute and drive has a "parking fee" attribute.  You want to know whether you could measure an interaction between "wait time" and "parking fee"?

I don't think that makes a lot of sense, since there is no alternative whose overall utility the two attributes both contribute to.  

That said, you CAN model the effect of parking fee on taking the bus and you can model the effect of wait time on driving - these are cross-effects.  It takes a bit of manual coding of the design matrix, and you'd need to run this in our standalone latent class or HB analysis programs, not in Lighthouse Studio, but it's a sensible thing to do and I've built plenty of cross-effects models like this.
answered Apr 20, 2020 by Keith Chrzan Platinum Sawtooth Software, Inc. (105,750 points)
I'm thinking about food-related choices between two very specific types of food and want alternative specific constants to make the choice options more realistic.  

This is a made-up example, but let's say sugar content is an attribute of interest. It could be a "common" attribute, but the realistic ranges for alternative A and the ranges in alternative B vary greatly.

So for alternative A, I might want to use  10/15/20g as levels. And for alternative B, I might want to use 30/45/60g as levels. It's technically the same attribute, but I want to describe it differently under different alternative specific contexts.

I want to know if lower sugar content in A: 1) increases the probability of B being selected and 2) increases the utility of sugar content in B (relative to when the sugar content in A is high.

Is there a different way to think about this? Should I use sugar a common attribute with restrictions?

Are standalone latent class or HB analysis programs something you purchase to gain access to?
I think the first part (1) makes perfect sense, and is easily doable using cross-effects.  I'm not sure how to model (2) and I think it would be a complicated model for which you'd need to do a lot of customization.  I would not go that route.
Ok thanks. Are the "standalone latent class or HB analysis programs" something you have to purchase?
If you're trying to do this using our software, then yes, you would need to buy some of our software (I think if you are already a subscriber to our Lighthouse Studio product then you can get these standalone products for free, but I don't really get involved in software sales, so I'd encourage you to check with our Support group).

But you can use any general purpose logit programs, even free ones in R, to run cross-effects models.
To get at (2), do you think I could run three different tasks:

A 10g (fixed) vs B 30/45/60g
A 15g (fixed) vs B 30/45/60g
A 20g (fixed) g vs B 30/45/60g

and then compare the B sugar utilities across those sets?
I am not sure what this analysis would tell you, are you?
Let me modify the example to help it make more sense.

Sugar content levels for alternative A: 5/50g
For alternative B: 0/30/60g

A is generally preferred and typically available with 50g in the real world. But the goal is to phase out A, and promote B.

I want to get at demonstrating the full interaction we expect. When sugar in A is 5, B is more likely to be chosen (than when A is 50); but only when or particularly when the sugar content level in B is 30 or 60.

When sugar in A is 50, the sugar in B doesn't really matter to anyone. Maybe fiber content is the most appealing part of B. But when A changes fundamentally in sweetness with this new low sugar content, the sugar content in B might become more important and more paid attention to. So to get sales of B up, sugar has to go down in A but be available at higher levels in B.

Is there a way to demonstrate something like that?
I would still model it with a cross-effect.  The sugar content of A affecting the choice of B.  

As noted above, I don't know a simple model that would allow you to model the sugar content of A affecting the utility of sugar for B.
...