Framing the simulation - Willingness-to-pay for features

Question

Hi everyone,

I am conducting a conjoint analysis on heat pump features in order to calculate the WTP of every (this is important) feature included in the survey.

I would like to do it through using either CBC or ACBC HB and reaching a small number of respondents (50-150 max probably).

My list has these attributes and levels (in brief):
SmartConnectivity: With/Without
Noise reduction: Low noise (-5dBA)/Standard (60dBA)
Efficiency: 750EUR yearly cost / 850EUR yearly cost
Tank size: 190L / 230L
Reliability: Deficiency reliability / Lacks deficiency reliability
HMI: Colour display and touch HMI/Simple HMI
Price: 3500/4500/5500/6500 (if CBC)
For ACBC I have an average WTP for each feature added and price as sum-of-attributes.

My questions:

1) Calculating the WTP for each feature through market simulation (two products side-by-side - one with the feature other without and a none included) and changing price of the one with the feature until it reaches the same market share as before.  Holding these two  products identical, will the other features presence - or lack of - influence my result?
And at what price for the base case scenario should the simulation begin with? Does it matter?

2) I do not think I will include other brands in the survey. So another option for the simulation would be to include other product options found in the market with their respective prices. Would that work without brand included?
Here I might run into problem with defining random levels for the other features of the test product, correct? How to solve it?

3) Is it crucial to define levels as best/worst and low/high?

4) Given the scenario of >6 attributes and <100 respondents and the goal, should I even consider CBC?

Also, please do share your suggestions and feedbacks since I am a beginner and could make use of those. I am also willing to know how you would do it :)

Answer 1

Generally, the two-product 50/50 simulation approach (with and without product enhancement) overstates the WTP.  We would recommend running simulations with a more realistic set of competitors and the None alternative (say, 5 or so total products plus the None).  Then, simulate the WTP product (the product for which we'll be estimating WTP) without the product enhancement.  Write down its share of preference (perhaps it is 15%).  Now, enhance that WTP product with a change to one of its attribute levels.  Next, find the change in price that drives the share back to the target original amount (15% in this example).  This approach gives a more realistic and lower WTP than the 2-product approach.

And, yes, it does matter what price you start simulating the base case at for the WTP product.  And, it also matters for your 50/50 two product case (unless you've fit a purely linear term to the price attribute across its continuum).

Its typical to have 200 or more respondents for CBC, when you are trying to project to universes of 500 or more consumers.  So, 100 is a bit small.  But, your results are probably much better than a guess (without any data), so there could be value even with small sample sizes.

Comments

Hi Bryan,
Thank you for the feedback.

Regarding that:
1) Since the initial price matters, how should I decide which price to begin the simulation with?

2) Since I want the WTP for every single attribute/feature and not a certain whole product, I would do an one-by-one change on attribute and leaving the rest of the features not currently analyzed assigned at random levels. Can that influence my results negatively? If so, how can I overcome it?

---

OK, you are a clear future beneficiary of the WTP feature we are building into our market simulator.  That will take us a few more months to release.  In situations where researchers don't want to assume a specific product or a specific set of competitors, it will run thousands of loops over different starting points for all the attributes as well as different draws of random competitors.  That way, it will come up with a "generalized" WTP estimation for attribute levels rather than WTP that can depend quite a lot on the starting assumptions.   The 2-product 50/50 approach makes it less dependent on starting point (though it still is dependent on price starting point for normal part-worth utility functions).  But, again, the 50/50 two product simulation approach tends to overstate WTP.

---

That is great to know, sounds like a fine (and complex) simulation.
OK, so I will go  with the 2-product 50/50  approach and have a disclaimer on it. Otherwise I might check the possibility of having specific products.
Appreciate your timely answers, they were helpful.