How many choice tasks for an MBC?

Question

We are thinking of doing an MBC.  This will be our first time.  We are experienced with CBC and comfortable there.  However, we aren't sure if there are rules of thumb that differ for MBC with regard to how many tasks to show per respondent.  
For example, lets say I have a list of 20 alacarte  items and 7 bundles.  Would we present experimentally designed mix of alacarte and bundles a series of times?  There would be no need for price variation...prices for each item are fixed.  How many tasks are recommended?  We want to not tax the respondent?
Thanks!

Answer 1

We do usually see MBC surveys give each respondent multiple tasks and in fact you must do so if you plan to use HB to run your analysis.  

How many tasks depends on how many items you have, how many you expect to be chosen in each task, how many respondents you'll have and how many predictor variables in your various models.  The most widely referenced rule of thumb for binary logit models (which is what folks often use to analyze MBC experiments) is that your number of respondents (n) times your number of tasks per respondent (t) should equal or exceed 10 times the number of parameters (k) in your model divided by the lesser of the probability the item will be chosen or that it will not be chosen (p):

nt = 10 k / p

So as the number of parameters goes up or the probability of the item your modeling being chosen goes down your number of observations n*t should rise.  Please email me (keith@sawtoothsoftware.com) if you want more details or a reference to the authors of that rule of thumb.

Answer 2

Keith Chrzan provided a more scientific answer than I will here.

From a practical standpoint, respondents can often be quite repetitious in their answers to MBC questionnaires.  Not surprisingly, they want very similar combinations of things across menu tasks, and often even substantial price changes won't dissuade them much from their choice patterns.  

A recent simulation study by Neuerburg at our last conference suggests that results don't improve that much for MBC under HB modeling once respondents answer 5 or 6 choice tasks.  I tend to favor about 6 to 8 MBC tasks for my work so far.  But, MBC encompasses so many kinds of designs and situations, that I suspect both Neuerburg and my opinions won't hold for all situations.

Comments

Oh, I should add that in your specific situation it sounds like you aren't doing a typical menu where the items are always there and only the prices vary.  You seem to be planning a type of study where the main issue is availability (items and bundles appearing and disappearing from the menu).  The answer regarding how many tasks you should do may be different (and more) in your case because of the greater variety of the menu composition across tasks!