Choice Design for Mode Choice

[#1]

Hi everyone,

I’m relatively new to SAS JMP, and I’m currently trying to design a discrete choice experiment (DCE) for my research. I need to create a set of choice scenarios where respondents choose between four transportation modes: car, motorcycle, bus, and train. Each mode has different attributes and levels, as shown below:

1. Car

• Travel time: 40, 60, 80

• Travel cost: 45,000; 55,000; 75,000

2. Motorcycle

• Travel time: 40, 50, 60

• Travel cost: 25,000; 30,000; 40,000

3. Bus

• Travel time: 60, 75, 90

• Travel cost: 2,000; 4,900

• Waiting time: 5, 10, 15

• Access time: 5, 10, 15

• Access cost: 6,000; 8,000; 10,000

• Parking: 0; 5,000; 10,000

4. Train

• Travel time: 25, 35, 45

• Travel cost: 3,000; 5,000; 8,000

• Waiting time: 5, 10, 15

• Access time: 5, 10, 15

• Access cost: 6,000; 8,000; 10,000

• Parking: 0; 5,000; 10,000

As you can see, the alternatives do not share the same attributes. Private modes (car, motorcycle) have no waiting time or access time, while public transport modes (bus, train) include these attributes. Some attributes, such as parking cost, apply both to private and public modes but with different roles.

I would like to generate:

• A D-efficient design

• A total of 16 choice tasks divided into 2 blocks

• Each respondent will only receive one block

My main questions:

1. Is it possible to build a choice experiment in JMP when alternatives have different attributes (non-symmetric design)?

2. How can I correctly define attributes that apply only to certain alternatives (e.g., waiting time only for bus/train)?

3. Does the JMP 19 Trial version limit any Choice Design features?

4. What are the exact step-by-step procedures to input this type of mixed-attribute design into JMP so that I can generate a valid D-efficient choice set?

I’ve explored the “Choice Design” module, but it appears to only support alternatives with identical attribute structures. I haven’t found a way to specify alternative-specific attributes like in my case.

Any guidance or examples would be greatly appreciated.

Thank you!

Kind regards,
Adam

[#2]

Hi @Bravoadam18,

Welcome in the Community !

It may be hard to comment on your topic without having more background context and informations, I however have some questions regarding your experimental design setup:

• What is your objective with this survey ? What are the goals of your research ?
• Why are the levels of the different transportation mode different ? I find the choice to make quite difficult, particularly when options are very close to each other, for example : Car 40min vs. Train 35/45min ? And I fear the comparison between very different choices may lead to obvious conclusions : I don't think someone would choose the option bus 90min when facing another alternative with significantly lower travel time (<60min).
• Maybe some factors and levels could be changed, as they represent the same idea behind :
  • For example, why would you split the waiting time from the travel time ? You could have one factor "Travel time" with different levels, and including both travel and waiting time.
  • For example, why would you split the access cost from any other cost (like parking) ? You could have one factor "Travel cost" with different levels, including the total cost of the travel (fuel, parking, ticket, ...) 

I don't know what is your goal behind this survey, but I would probably group some factors and levels together to have less options in order to have a more simple but more reliable model and explanations/interpretations. Maybe the design could be done by using different factors and levels based on different criteria :

• A factor related to travel time with levels : <30min, between 30 and 60min, more than 60min
• A factor related to travel cost with levels : low, medium, high (to be defined with appropriate cost ranges)
• A factor related to travel comfort with levels : low, average, high (to be more explicitly defined)
• A factor related to environmental impact, with levels : low, medium, high (to be precised with Co2 footprint ranges or similar indicators)
• ...

This setup could help understand what are the key priorities of transportation user and their main criteria for choosing their transportation mode (time, cost, comfort, environmental impact). You may also be able to cluster respondents based on their responses and the impact of the different factors.

Choice Designs are helpful to understand what are the key levers/drivers that influence the choice of consumers. Analyzing the several selected options picked in different choice sets enable to evaluate the relative importance and significance of the attributes (factors) tested, and find the best level combinations in terms of utility score across all respondant responses. You can find an example here : Run the Choice Design and Analyze the Results (jmp.com) If you use different factors and many different levels, I fear that you may not be able to conclude easily. There are still options that could allow you to do that :

1. You could create the design using the Custom Design platform with arbitrary levels 1/2/3, the modify in the generated table these values by the "real" values, and use a nested model and/or Choice Models to analyze the results.
2. In case you still have very diverse options you want to include in the design that would be very tedious or impossible to create with various methods Restrict Factor Level Combinations, you can create a Candidate set with all possible combinations you want to create your design from, and use the Custom Design platform to use the runs from this candidate set to generate a design (through the option Select covariate factors. More info in the presentation : Candidate Set Designs: Tailoring DOE Constraints to the Problem (2021-EU-30MP-784) 
   I think this option may be relevant here, as some "extreme" combinations may be completely unrealistic.

You can also watch this presentation to learn more about Choice designs : Case Studies on Designing and Analysing Discrete Choice Experiments Using JMP® - JMP User Community

Hope this first discussion starter might help you and may help other JMP users to join the discussion,

[#3]

Hi @Victor_G,

Thank you for the detailed feedback. I realize my initial explanation might have been a bit incomplete regarding the background of the study.

To clarify, the primary objective of this research is to analyze individual preferences in modal choice. Ultimately, I aim to determine the probability of an individual choosing a specific mode (mode share) and to identify the significant factors influencing those decisions.

Regarding your questions on the design:

1. Varying Levels & Elasticity: The attribute levels are intentionally varied to measure individual elasticity (sensitivity) when facing increases or decreases in each factor. For instance, the travel time levels reflect different real-world traffic conditions (congested, normal, and free-flow) for private vehicles and buses. While a 90-minute bus ride might seem like an unlikely choice compared to faster alternatives, the bus fare is set as the lowest among all modes. This is designed to capture the trade-offs made by specific demographics (e.g., lower-income groups) who may prioritize lower costs over travel time.

2. Separating Time & Cost Components: Regarding the separation of time components (travel time, waiting time, access time) and costs: Since two of the alternatives are public transport, previous transport literature suggests that these distinct components carry different weights in a user's utility function. Therefore, it is crucial to keep them as separate attributes rather than aggregating them into a single "Total Time" or "Total Cost" factor.

In the final questionnaire, respondents will be presented with random combinations of these attributes (Choice Sets) and asked to choose one of the four alternative modes. Each respondent will face, for example, 4 different choice scenarios.

Thank you for the guidance, I really appreciate it. The suggestion regarding "Custom Design" seems very applicable to my specific case, and I will definitely look into it further.

Thanks again for your time.

[#4]

Very interesting, thanks a lot for the explanation @Bravoadam18, it's a lot clearer.

I think the Candidate set option and the use of Custom Design could help you generate an appropriate survey plan: The addition of Block factors can help you define the number of profile per choice set, while the use of the option Group runs into random blocks of size in Design Generation can help you distribute the different choice sets in the two surveys (number of choice sets per survey): Choice Design Terminology

Hope you'll find some ideas with these suggestions,