Is a randomized design a viable option for choice-based conjoint and latent class analysis?

Question

I need some advice re/ randomized designs in CBC:
- In my CBC, I have n = 220 and 10 pair-wise comparisons, i.e., each respondent evaluates 20 choice options in 10 choice tasks. Each choice options has 5 attributes á 3 levels.
- In my pre-study, I used a d-optimal fractional design generated, i.e., each respondent saw the same 20 options.
- In my main study, my survey tool was accidentally set to random design, i.e., each respondent saw 20 randomly generated options.

I am not sure if there problems associated with a random design - at least all CBC articles I came across use fixed designs.
Could you help me out here, as I need to decide whether I can proceed working with the conjoint data obtained from random design, or whether I should repeat the study with a fixed design.

Many thanks for your help!

Answer 1

A more efficient design is sometimes a little bit better for some purposes, but I've used the Random design in our designer for some LC applications in the past.  I don't see a problem.

Comments

Thank you so much for the quick response, Keith!

So just to make sure: Basically each of the 220 respondents had a different, random set of 10 pair-wise comparisons with randomly generated choice options (not: a purely randomly generated set of 20 choice options that was the same for all respondents).

But there is no problem in that? (I am just investigating main effects)

Again, thanks for your help!

---

No, no problem that the versions are different - that just means you covered more of the experimental design space, which isn't a bad thing.  
 
Of course if you only care about main effects your experiment would have been a little more efficient (smaller standard errors) if you had chosen the Complete Enumeration or Balanced Overlap strategies in our software - but if you run some tests, you'll see that your Random design isn't a whole lot less efficient.