choice-metrics.com

Dear all,
I'm trying to develop an experimental design for a stated-preference survey which I'm not sure can still be called a choice experiment.
The survey is about preference for land use change. Alternatives are future use for an abandonned piece of land. They are characterized by only 2 attributes: future land use and price.

The main assumption is that preference for land use change depends on attributes of the piece of land (status quo): vegetation state, location and land fertility. These attributes can each take 2 to 3 levels. I intend to estimate preferences for land use change depending on these attribute levels of the status quo. In other words, I intend to estimate the interaction between the attributes of the status quo and the preference for land use change.

I cannot do a full-factorial design since the number of combinations of SQ attribute levels are two high (about 80). I would therefore like to use a fractional design. I don't know if this is a design that can be generated using Ngene. I also do not know which model to use to analyze this problem. I could not find similar problem/question in the literature.

If somebody as a clue or a lead on how to deal with this problem it would be highly welcome.

Best,

I believe that you want a design with varying scenarios. A choice task would look something like this:

Consider a piece of land with the following characteristics:
* Vegetation state = ...
* Location = ...
* Land fertility = ...

What would you choose to do with this piece of land?

Option A (status quo): Nothing, Cost = 0
Option B: Develop as ..., Cost = ...
Option C: Change to ..., Cost = ...

In other words, vegetation state, location, and land fertility are scenario variables that are constant across all alternatives but vary across choice tasks, while the choice alternatives have attributes land use and cost/price. In the utility functions, you would need to interact the attributes in options B and C with the scenario variables. Since these are mostly categorical, you would get a lot of interactions.

It would be simpler if you would had two labelled alternatives:

Option A: Do not develop, Cost = 0
Option B: Develop the land, land use = ..., Cost = 0

In this case, you could add the scenario variables directly as main effect into the utility functions.

If the types of land use are limited, it would be even better to use:

Option A: Do not develop, Cost = ...
Option B: Develop land for use X, Cost = ...
Option C: Develop land for use Y, Cost = ...
Option D: Develop land for use Z, Cost = ...

In this last case, only cost varies as an attribute, while alternative-specific constants are estimated for the specific land use options, and the scenario variables can enter the utility functions as main effect.

Michiel

Thanks a lot Michiel for understanding and proposing concrete solutions to my peculiar question. I like a lot your second option in which I would have 8 scenarios and 4 land use options.
The only alternative would therefore be the payment.
If we take for example 8 level of payment and a number of 8 choice sets to cover the 8 scenarios, do you think Ngene can help me in setting the design for the payment levels to be affected for each alternative in the 8 choice sets.
Thanks again

Yes Ngene can create a design for each of the options I mentioned.