Have an idea?

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

CBC WTP Scenario Error "Extrapolation..."

With my data including a conditional pricing and interaction effect (both on attribute "Productlevel"x"pricelevel" I try to run a very simple WTP scenario.

There is an error indicating:  "Extrapolation for attributes with interactions is ot permitted"

Manually I can find a price that equalizes share of choice for an improved product .

a) what could be the problem?
b) is it wise or not to use constraints for HB estimation  for price and/  or productlevel here? (it is clear that product-level has a low-mid-high preference)

Thank you
asked May 24 by steve

1 Answer

0 votes
Sorry about this, the simulator doesn't allow extrapolation for an attribute involved in an interaction effect.  See explanation in the final paragraph of our documentation here:

So, for some of the draws of your WTP scenarios, it's having to try to extrapolate to share-equalize the test product.

A possible solution is to estimate a new utility model for the purposes of conducting WTP analysis and to not estimate interaction effects.
answered May 24 by Bryan Orme Platinum Sawtooth Software, Inc. (199,115 points)
Thank you.

So - do I need an interaction effect, or can I leave it? If it also works without and gives me my WTP - separately for my three products on the respective price levels?

To better understand:

- I guess in any way, it makes sense to constrain price here "high-to-low."
- Correct that constraining the product level does not make sense with conditional pricing? So it is ok that "better" products get lower part-worth values as they are connected to the higher prices?
- If it was an alternative-specific design: Does WTP not work there? (as it always includes interaction?)

I hope my questions make sense.
Indeed, if the "product level" triggers conditional pricing and one product (say a premium one) on average shows at higher prices than a discount product, then it doesn't make any sense to constrain the utility of the premium product compared to the discount one.  With conditional pricing, we no longer can interpret the utilities as being "all else held constant"...because NOT all else was held constant (price depended on the product levels!).

If you've estimated individual-level utilities with HB, the need to estimate interaction effects to explain further variance is typically much lower than if running pooled (aggregate) models.  Then, our software is limited in its ability to extrapolate beyond price ranges when interaction terms are involved.  So, that's another reason to drop the interaction term if wanting to use our software for its WTP capabilities.

Alt-spec price attributes should work fine...as you typically do not ask the software to layer an interaction effect on top of an alternative-specific effect (as the alt-spec price effect is already automatically "interacted"...conditional upon.. the other attribute level that triggered it).
Thank You