n many of our projects we need to leverage the ASD capability of Lighthouse where certain attributes only apply in the presence of specific levels of a primary attribute. Example: A retailer may consider various means to boost sales. The primary attribute may be "sales promotions" and one of the levels may reflect the idea of a customer card that when presented gives the customer discounts. A secondary attribute may then determine the discount amount from "-2%" to "-20%".
The challenge/issue we struggle with is that the levels of the secondary attribute are zero-centered which frequently leads to the following undesirable dynamic:
Let's say the primary level "customer card w/ discounts" is always going to have a net positive impact/lift over "no sales promotion" level. The least favorable specific discount amount, however, will necessarily always have a negative utility. Often the negative utils of "-2%" (in our example) can overcompensate and negate the lift of having a discount card - turning an objectively valuable promotion option into something that can reduce adoption compared to not doing anything.
The only remedy I can think of - aside from manipulating the utilities to avoid net-negative simulation impact - is to recode the experiment response file in a way so that each promo option get's a dedicated attribute with the levels (for the discount card) "no card" "card with -2%" "card with -5%" "..." etc. However, this may introduce correlation between "no card" and some other promotion option being present - which then can result in "no card" being assigned the positive impact of the other option and similar confounding of cause-effect dynamics by the estimation.
This is a long shot but I'm hoping that folks in the community can share their preferred strategies and work-arounds how to deal with this situation.