Does the choice simulator take an individual respondent perspective when estimating shares of preference?

Question

Hi everyone,

I have completed an ACBC survey of 164 respondents and am experimenting with the choice simulator.

When using the shares of preference model, does the simulator individually predict the response (acceptable or not) for each survey respondent to a stated concept based on their estimated level utilities?

i.e a 50% share of preference would suggest that 82 of my respondents were each individually predicted to have found that concept acceptable?

Thanks very much!
James Durrand

Answer 1

Hi James,

Correct,  the HB analysis creates an individual model for each respondent.

So when you create a simulation, the model runs this for each respondent and reports the aggregate results

If you were using a first choice approach to the modelling then it could be interpreted in the way you mention, but with share of preference its a touch more complicated.

This is because one respondent could have a 25% share of preference for your simulated product and a 75% preference for none.

If we had two respondents the other could have a 75% share of preference for the product and only 25% for none.

The average of these two respondents would be 50% share of preference, but in reality only one respondent would take the product over none.

So a share of preference is more easily thought of as the probability of the sample choosing the product, rather than 50% of your sample explicitly chose the product.

Comments

Hi Dean,

Thank you for your prompt response that is very helpful.  

I have re-run the simulation with a randomised first choice model and the estimated shares of preference are similar under this. It is also likely to be more intuitive to understand for the intended audience!

I appreciate this model assumes the respondent chooses the concept with the highest summed utility (for them individually) based on their individual level utilities.  If a 'none' option is inlcuded, how is the summed utility of the 'none' alternative estimated for each respondent?

---

There is no summed utility for none - none has its own utility.  

In an ACBC this is initially estimated via the screener section using the "not a possibility for me" selections to gauge its value.

If you used the calibration questions you could also re-estimate the none parameter with your chosen threshold too.