Have an idea?

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

Applying None weight in calculations

Dear colleagues,

Could you explain how do you calculate in SMRT share of preference for None option using None weight? I used the following approaches in Excel:
1. =exp(NoneWeight*NoneUtility)/SumOfExponents
2. =NoneWeight*exp(NoneUtility)/SumOfExponents

I got 100% sum of shares but in both cases I did not get the same numbers as in SMRT.


BR,
Alexey
asked Jun 27, 2017 by Alexey

1 Answer

0 votes
The none weight in SMRT is applied as NoneWeight*exp(NoneUtility), which results in its non-normalized share.  This is done before the SumOfExponents is calculated, which is used to normalize the shares to their 0-100 form.
answered Jun 27, 2017 by Walter Williams Gold Sawtooth Software, Inc. (17,000 points)
Dear Walter,

Many thanks for rapid response.
I will double check my calculations.

Kind regards,
Alexey
Hi,

I have had the same problem. When I build my simulator, I have one set of shares. However, when I attribute the weight 1 to the none option, I seem to have problems to match the shares. How to proceed in such cases?

Thanks a lot for your time.

Regards,
Are you building your own simulator, and comparing it to the SMRT simulator?  Are you using Share of Preference?  Can you provide more detail?
Hello, Walter,

Yes, I am building my own simulator. I am using Share of Preference in the calculations. I have seven price levels (1 to 7), and my "Market Price" is the level 4. When I add the NoneUtility to 4*PriceUtility, I match my market share. However, as I change the price levels, the shares do not match.
In our estimation software (assuming you used HB), if you coded your price attribute as Linear, then the level values (1-7) would be zero-centered in the design matrix (-3 to 3).  When doing simulation, the values should also be zero-centered before multiplying by the coefficient.
...