Have an idea?

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

Max product alternatives in simulation?

From the SSI Web Help "None" Weight: "With individual-level modeling under CBC/HB, we have seen that the share of "None" is approximately correct irrespective of the number of products used in the simulator (within a reasonable range of product alternatives)".
In your experience, what is a "reasonable" range of product alternatives. In a portfolio optimization, the number of SKUs can be quite large (e.g. 40 to 50), what is your experience in dealing with such cases?
Another related question: how do you typically ensure that the model does not automatically yield a lower share for a brand as its number of SKUs is reduced? SKU reduction may actually be a good a way of simplifying/easing the buying process for many consumers.
asked Sep 23, 2013 by pappe

1 Answer

0 votes
If I remember correctly, when we looked at the None weight for CBC/HB modeling, we only looked at varying the number of products in the simulation by maybe one or two from what was shown to respondents.  So, if you showed respondents 5 products per screen, plus the none, you might have reasonably good predictions of the None share when going to 4 or 6 products per screen.

Now, if you are doing a CBC with 40 to 50 SKUs per screen, then maybe you could vary the number of simulated products by a few more and still get pretty accurate predictions of the None%.  But, it probably depends on the particular setup and which products you are adding or deleting from the scenarios.  The best rule of thumb is to only predict the None% for scenarios with the same complexity (number of concepts) that you've actually shown respondents.

So, the only way to know this for sure would be to add some CBC questions that used different numbers of product alternatives--then you could observe the None% given and compare to the predicted None% from your model.

As a side note: Please be very careful with using more than about 20 product concepts in the market simulator with RFC (randomized first choice).  The number of iterations should be increased from the default 200K iterations to something like 500K or even 1MM.  That's because the RFC process purposefully adds some noise to the shares (that's how it corrects for product similarity), which noise smoothes out over thousands of iterations.   But, if the typical shares are really small (such as <5%), then you need to ask the algorithm to do more iterations than the default so that the random error added isn't too large relative to the size of the final predicted shares.
answered Sep 23, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (134,015 points)
Thanks Bryan - I am actually thinking about fewer product concepts (up to 10) and 5-6 attributes in the design/collection. This is to measure the sensitivity to these attributes and keep the exercise manageable to the respondents. In this case, it seems like you advice is to limit the number of product alternatives in the simulation to about 10 or 12. How about simulating 10 to 12 attractive offers and recording the share None and keep the latter constant in subsequent simulations involving a larger number of product alternatives? This way, the simulator does not artificially allocate the share of people who would not buy to the additional products. Let me know what you think. Thanks.