# Conditional pricing for one concept?

I need to add a conditional pricing for a certain combination of attributes and levels, i.e. to one precise concept (not only one or two attributes), where the price attribute must be replaced by a customized price. How to proceed without replacing all the price attributes of all possible concepts?
I don't think I fully understand your question, but something about it makes me nervous.  If your intent is to show a specific price for each combination of attribute levels that makes up a concept (and price does not vary randomly), then the total price of that product concept will be perfectly predictable by a linear combination of the attribute levels that make it up.  That would be a case of colinearity in the experimental design and would be a confounded experiment.  You would not be able to separate the effect of price from the effect of the other attributes.  Maybe I'm jumping to conclusions from your question, but it was making me worried.
Bryan, many thanks for answering. Sorry for making you nervous.
I will give you an example: my CBC design is set up with 2 concepts for task + opt-out alternative, 5 attributes per concept including price, 2 or 3 levels per attribute, depending on the attribute. As I foresee a big sampling (2000 interviews) I choose the random tasks and a full traditional CBC profile.
One and only one concept (consisting of A1L1, A2L1, A3L2, A4L1, let's call it concept Bob) has a fixed price, determined a priori. All others must randomly vary. Proceeding with the conditional pricing feature, how can I limit my price replacement only to Bob?

OK, this will not foul up the experiment, but it will cause some headaches for you to do proper analysis.  Explanation below.

Let's imagine you have 5 attributes, with levels as follows for the five attributes 2x3x2x3x3, where we assume price is the last attribute with 3 levels.

That means there are 2x3x2x3x3=108 possible combinations to show respondents.  Approximately 1 out of 36 combinations shown to respondent will be of the magic "bob" concept, where you have to fix the price.

So, you should use conditional pricing (n-way pricing) where price is conditional on the combinations of the 4 other attributes.  I'm assuming you'll be letting price just take on 3 values (the main effect of price) for all of the 108/3-1=35 product combinations.  But, when "bob" concept appears, it will always show the same price for all three price levels.  Go ahead and fill out the 108-cell conditional pricing table to reflect that fact.

Now, the headache comes during estimation, where you will need to do some custom coding of the data file and will need to fit a linear or log-linear term to price to fit the data correctly.

You'll need to export the CBC data to a .CSV format.  From Data Management area within Lighthouse Studio, add a Job (Add Job...).  Select Add button, and click "CBC (.csv, .cho)" format.  Under File Format on that same dialog, click "Single Format .CSV".  Export the job.

Now, open the .CSV file that was exported from the above in Excel.  You'll see a column for the Price variable which currently takes on 3 states (1, 2, or 3).  You'll need to write a few formulas in Excel to replace price with the actual price shown to respondents for each concept.  And, when concept "bob" is shown, you'll need to fill in that constant price.

Save the modified .CSV file so that the Price column now contains the actual prices shown to respondents rather than 1, 2, or 3.  (See more info below regarding scaling these prices for proper HB convergence!)

Now, I'm assuming you'll be doing HB estimation, so you'll need to estimate the part-worth utilities using the standalone CBC/HB Module (rather than Lighthouse Studio, which cannot handle "user-specified" prices) that you can download from: http://www.sawtoothsoftware.com/support/downloads/download-cbc-hb

But, prior to using the standalone CBC/HB Module to estimate the part-worth utilities, you'll want to make sure you have scaled the prices that you inserted to lead to proper convergence in HB.  We have found that the prices should not exceed a range of about 1 to 10 to lead to proper convergence.  So, if your prices shown to respondents ranged from 5000 to 10000, then you should divide all the prices in the Price variable column of the data file by 1000, leading to prices ranging from 5 to 10.  Decimal places of precision are supported, of course.

The procedure above will lead to a linear estimation for price once you've set up the HB run properly in the standlone CBC/HB module.  To do this, open the modified .CSV file in the CBC/HB Module.  On the Attribute Information tab, click the drop-down control to change the Price attribute from part-worth to "User-specified".  (This might already have been done for you by the system, when it recognized the decimal places of precision in the data file for the Price attribute.)

Be careful when you read the part-worth utilities plus the linear price coefficient into one of our market simulators that the market simulator recognizes the right price range that you used.  You'll need to indicate to the simulator that the Price variable is a continuous variable.  If you divided the prices by 1000 (for example) for estimation, when you specify the prices in the market simulator you will be specifying prices divided by 1000, or else the price coefficient you end up applying in the market simulator will be 1000x too sensitive.
answered Sep 28, 2018 by Platinum (172,790 points)
Many many thanks. Let's go for the headaches
Bryan, sorry for bothering again on that: what if instead of adding one conditional price (let's say 20€) to the one concept (Bob) I consider that price as a further level of my price attribute (2€, 5€, 10€, 15€ AND 20€) and I eventually define a prohibition that avoids the coupling of 20€ to all concepts except Bob? Would that simplify my job?
If your conditional pricing adjustment does something funny based on a trigger involving more than two attributes, then you run the serious likelihood of a bad fitting and bad predicting model.  Main effects and two-way interaction effects cannot account for it.  So, in short, you still would be facing the extra work I earlier described for building a model that properly accounts for the profiles and prices shown to respondents.
Thanks Bryan. I am afraid I wasn't clear. I don't mean to involve more attributes in my conditional pricing. The idea is not to set any conditional pricing at all. I would rather add one more level  to the price attribute in my design and eventually set a prohibition for the fourth price level with all the concepts except the desired one.  First time for me with such a big and important survey, sorry.
In that case, try the prohibition and then do the standard Test Design approach.  You are hoping for standard errors for your main effect utilities of 0.05 or less.  And, standard errors for your interaction effects of 0.05 or less.  Also, recognize that if you specify a prohibition between an attribute and price, you will no longer have the ability (without doing power tricks) to estimate an interaction effect between those two attributes.

One possible solution, if any of your standard errors turn out poorly under your new strategy, is to create two price attributes and to use alternative-specific designs.  There is a section in the CBC Help manual that describes alternative-specific designs.