Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Standard Errors In partial Profile CBC


I am wondering if there is a rule of thumb for the size of the standard errors in partial profile designs, in the same way that there is for Alt-specific designs.

Are the attributes which are rotated through partially (seen in one choice set and assumed to be held constant across alternatives but unseen in another choice set) susceptible to having slightly higher SEs than the non-rotated/constantly shown attributes attributes. i.e would a SE level of 0.1 be acceptable or should the 0.05 error level still be the desired threshold for all attributes?

Many thanks
asked Apr 23, 2017 by Jasha Bowe Bronze (1,680 points)

1 Answer

+2 votes
Best answer

I probably would allow the partial profile designs to have larger standard errors, just like interactions and alternative specific designs.  My reasoning would be a bit different, though.  Whenever we've compared partial profile to full profile stimuli, we lose something in terms of statistical precision (i.e. the partial profile models have about twice the amount of statistical error that full profile models have).  Counteracting that, however, is the fact that partial profile models have much less response error than do full profile models (it varies a bit from study to study, but on average, partial profile designs produce only about a quarter as much response error as full profile designs).  So in terms of net error, partial profile designs are giving you better efficiency than you might think if you just looked at statistical precision (standard errors).
answered Apr 23, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (60,075 points)
selected Apr 24, 2017 by Jasha Bowe
Thanks Keith, what a great answer... It was my feeling that the SEs might be slightly larger but I didn't consider the extra efficiency that would be gained in the  reduction in response error, but this makes perfect sense as the whole purpose of partial-profile designs is to make long attribute lists shorter for each choice task and thus easier to engage with/respond to.

Thanks again