May-09-2024, 02:03 PM
Background
I'm looking to implement a discrete choice experiment where participant(s) are presented with pairs of 'product profiles' and asked which product they prefer. Across trials different pairs of profiles are presented. Each profile contains a fixed set of characteristics (e.g. price, size & flavour), but with different combination of characteristic values (e.g. £2, Medium, Vanilla). From the outcome data one should be able to identify how the values of the characteristics influence the participant's choices, using something like Logit analysis.
I need to determine a design for the experiment (i.e. determine the number of trials, and which set of profiles to show on each trial). There is a significant 'Design of Experiment' (DOE) literature on this topic, and while there are standard 'orthogonal' methods of determining the DOE these tend to require increasingly large designs (i.e. large number of trials) as the number of characteristics or characteristic values increase. I need to keep the number of trials to a minimum. There are algorithms that provide 'adaptive' experimental designs. Basically the algorithm uses the participant's response from previous trials to determine what is the most efficient pair to show on the next trial. This can substantially reduce the number of trials required.
This is an example of an adaptive DOE algorithm: https://www.semanticscholar.org/paper/Pr...1d767332a9
Question:
Are there any python packages available which could be used for implementing an adaptive design (doesn't need to be the particular algorithm referenced above, just something similar)?
I've found some packages that provide DOE options (e.g. https://pypi.org/project/DoEgen/ & https://pypi.org/project/choicemodels/ ) but they appear to just use the orthogonal approaches.
I've found an adaptive algorithm ( https://pypi.org/project/adopy/ ) but it appears to only be designed for specific simple experimental designs relating to common experimental paradigms in Psychology).
Thanks
I'm looking to implement a discrete choice experiment where participant(s) are presented with pairs of 'product profiles' and asked which product they prefer. Across trials different pairs of profiles are presented. Each profile contains a fixed set of characteristics (e.g. price, size & flavour), but with different combination of characteristic values (e.g. £2, Medium, Vanilla). From the outcome data one should be able to identify how the values of the characteristics influence the participant's choices, using something like Logit analysis.
I need to determine a design for the experiment (i.e. determine the number of trials, and which set of profiles to show on each trial). There is a significant 'Design of Experiment' (DOE) literature on this topic, and while there are standard 'orthogonal' methods of determining the DOE these tend to require increasingly large designs (i.e. large number of trials) as the number of characteristics or characteristic values increase. I need to keep the number of trials to a minimum. There are algorithms that provide 'adaptive' experimental designs. Basically the algorithm uses the participant's response from previous trials to determine what is the most efficient pair to show on the next trial. This can substantially reduce the number of trials required.
This is an example of an adaptive DOE algorithm: https://www.semanticscholar.org/paper/Pr...1d767332a9
Question:
Are there any python packages available which could be used for implementing an adaptive design (doesn't need to be the particular algorithm referenced above, just something similar)?
I've found some packages that provide DOE options (e.g. https://pypi.org/project/DoEgen/ & https://pypi.org/project/choicemodels/ ) but they appear to just use the orthogonal approaches.
I've found an adaptive algorithm ( https://pypi.org/project/adopy/ ) but it appears to only be designed for specific simple experimental designs relating to common experimental paradigms in Psychology).
Thanks
