Sign Reversal in the Interpretation of Discrete Choice Experiments

[Maria Holmes]

Dear Forum Members,

I hope this message finds you well. My question might seem trivial or obvious to some, but I believe that a clear and definitive answer from the knowledgeable members here could help resolve my doubts.

I recently conducted a discrete choice experiment (DCE) with three attributes, each having three levels: low, middle, and high. My goal is to calculate the benefit of improving from low to middle and from middle to high. However, I am unsure whether I need to run two separate models or if one model would be enough. Here are the two options I am considering:

Option 1: Two Mixed Logit Models

• Model 1: Low level as the reference level, to calculate the benefit of the change from low to middle.
• Model 2: Middle level as the reference level, to calculate the benefit of the change from middle to high.

Option 2: One Mixed Logit Model

• Use the middle level as the reference level.
• For the change from low to middle, reverse the sign of the coefficient (since the model directly provides the benefit for middle to low).

My Questions:

• Is it valid to use Option 2?
• If yes, are there any specific considerations I should keep in mind when interpreting the results? I also want to calculate WTP for the changes and calculate interaction effects with other attributes.
• Does anyone know of a scientific source that explicitly describes this? Unfortunately, I haven't been able to find anything

Thank you in advance for your insights and advice! 😊

[George Ford]

Option 2 and use margins to get the differences of interest seems a sensible option, or lincom. You'll need margins anyway to transform the logit coefficients into something useful.