choice-metrics.com

Dear All,

As part of a lab-in-the-field study in East Africa, we plan to conduct a choice experiment. We will move from one village to another. We plan to visit 32 villages: 16 in Kenya and 16 in Uganda. Furthermore, we will have up to 24 participants per village.

My idea was to have a d-efficient design with 48 blocks of 8 choice sets. In the villages, we will then distribute either the first 24 blocks or the second 24 blocks. I just realized that we might have a problem if we do not reach 24 participants per village. Some blocks might be underrepresented in the overall sample due to our mechanism of distributing the blocks according to participant ID (which is randomly distributed between 1 and the max. number of participants per village).

Is it a problem to have some blocks less often represented than others?

Is it better to have a random allocation of the blocks? Or even fewer blocks?
However, here we might get the problem that there might be some participants within one village that actually play the same choice sets; some choice sets might be less likely to be played, and we have no control over this distribution of blocks.

It is a labeled choice experiment with 4 attributes with 2 to 5 attribute levels each.

Kind Regards
Philipp Händel

A design with 48 blocks of 8 choice tasks has 384 rows. Is there a reason why you would like to use such a large design? There is usually not much need to have hundreds of rows in a design, unless you have a very complex model with a large number of parameters.

I always try ensure that each block is completed at least once, but with so many blocks there is a high likelihood that some blocks are not completed at all. Also, it becomes more difficult to compare across different villages and countries if the completed choice tasks are very different due to certain blocks being unevenly completed. By reducing the number of blocks you would reduce this risk, although in practice you will always have some imbalance in blocks in the data set and that is fine. If you would like to compare villages, you may want to use the same blocks across all villages. For example, you could use 12 blocks of 8 choice tasks each, and use the same 12 blocks across all villages.

Michiel

Dear Michiel,

Thanks for the reply. This is very helpful. Then I will readjust our design and use less blocks.

The high number of blocks is basically driven by the idea that more blocks is always better and the expectation to get all 24 participants in each village. According to this assumption, we would have had an even distribution of completed blocks, at least at country level, but I wrote in the forum because we realized that we could not fulfil the assumption and therefore had to adapt our design.

Greetings
Philipp