Have an idea?

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

How to make summed pricing in ACBC depending on number of attributes?

I am currently setting up a conjoint with summed pricing. The product has in reality up to around 40 different features ('solutions' they call it), which are on/off. Each of these features have a price component, that should be added to the total price.
By a constructed list we limit the number of features to 10 or 12 relevant ones in the ACBC.
When adding up the price components of the different features, we could easily reach (or exceed) the total price of the machine.

I have 2 questions:
-how do I adjust the summed price, depending on the number of available features (on)? Could I e.g. work with a dummy variable in the price adjustment tab? If so, an example could be helpful
-at the analysis level, the product to simulate may contain up to all features, while the respondent will only be confronted with the relevant ones. My point is that this affects the total price; in the exercise prices could be lower than in the analysis. How do I make sure the prices are in the same range, still achieving a logical price sensitivity, even when having most features? I was thinking to not add any price component above a certain price (or a certain amount of features). And also assume that the respondent would want to pay for features that are not relevant.

Any advice? Do you have an efficient suggestion?

asked Jun 15, 2017 by andydevos (180 points)

1 Answer

0 votes
Sorry, there isn't any way built in to the software for the price adjustments to think in terms of how many attributes are present.  The only thing you can do is  set price adjustments for each level and set price adjustments for combinations of levels.  The adjustments for combinations table doesn't not let you specify a 0 level to indicate that the attribute is missing.

Keep in mind that it's probably not a great idea to run simulations outside of what you had respondents go through.  As an extreme example, if we showed people at most 2 features, it would be hard to convince me that a simulator is accurately predicting utility of a 20 feature option.

I'm not really sure what the best solution is to this, sorry.   Maybe someone else will have experience doing something like this.
answered Jun 15, 2017 by Brian McEwan Gold Sawtooth Software, Inc. (46,970 points)
Hi Brian,

Thanks for the answer. Unless someone else has a better idea, I will indeed try to cover this by price adjustments.

On the number of attributes: I am fully aware of the consequences of not simulating scenario's that are not covered by the exercises. However, that's what the client is aiming at. I have stressed this concern and will have to deal with it making rough assumptions on the features that were not seen by the respondent in the conjoint.

There's a paper from our 2013 proceedings where the author has a recommended approach for trying to do diminishing returns.  It's not exactly this situation since you are dropping things, but perhaps might introduce some ideas to help.    The link is http://www.sawtoothsoftware.com/download/techpap/2013Proceedings.pdf and the author is Kevin Lattery.
Hi all,

I have an additional question on a matter that complicates the analysis. We did use ACBC, but skipped BYO and screening (key parts in ACBC) and only keeping the tournament. The reason was that the BYO and screening were irrelevant in our context (I leave out the details). We made an a priori tailored selection of the features based  previous questions (scripted outside Lighthouse). So, the tournament does only include relevant products. And the main reason for using ACBC was the summed pricing ability.

The annoying aspect is that we should use our a prior selection questions to estimate the 'none'. One route was to reconstruct the screening tasks based on the a priori questions (or as proxy the missing attributes). How do I most efficiently approach this? I kind of know the structure of ACBC data file, but to manipulate it is not that evident.
Another option is to export it to a cho file and add tasks for the 'screening'. How can I then import such 'constructed' or 'augmented' file in Lighthouse?

How do I best continue this?

Modifying the raw data in an ACBC would be pretty awful.  You can augment choice data as you mentioned by exporting the data, and then adding additional choice tasks as if your previous questions were part of the conjoint data.  For example, you can add a choice task that shows brand A was preferred to brand B.

The downside is that you have to then use the standalone CBC/HB software, which outputs a single parameter for the pricing variable.  Inside Lighthouse we take that parameter and multiply it by the values you put in the analysis pricing table so the utility file looks nice and normal (i.e. you have a discrete utility for the high and low price point).