Eval command to re-block

[Lily820]

Hi,

This is my original design with 20 blocks with 10 choice sets in each block. Now we would like to reduce the number of choice sets in each block such as only five. Then we will need 40 blocks instead of 20. Can I just use eval command to achieve this? Will eval command generate the new choice sets with 40 blocks just as what mention in Ngene manual page 178? Or it just evaluate the new design? Thanks for your help.

Regards
Lily

Design
;alts = alt1, alt2
;rows = 200
;eff = (mnl,d)
;alg = mfederov(candidates = .....)
;block = 20

;model:
U(alt1) = ......

[Andrew Collins]

This can be done, the newblocking syntax is the key when used in conjunction with eval. Unfortunately there is an error in the Ngene manual (which we will fix for the next point release). The newblocking syntax is associated with the block command, not the eval command.

You can save the existing design, perhaps in the project. Then evaluate that design with the same syntax, except for two new or changed lines:

Code: Select all

;block = 40, newblocking
;eval = <design file name if in project, or full path of design file>

This will give you the same design levels, but newly blocked.

Andrew

[Andrew Collins]

Actually, the manual should be correct for almost all installations of Ngene 1.1.2.

[Lily820]

Thanks Andrew.

This works! Excellent! Thank you so much! Now I got 40 blocks instead of original 20 blocks within the original design.

May I ask one more question?

Since the original design is efficient not orthogonal, so blocking here actually cannot ensure attribute level balance right? So what is the point of re-block here? Just to reduce the number of choice sets in each block? For example, if we have two different samples, one is general population who will do the design with 20 blocks and the other one is patients who will do the design with 40 blocks. Are we able to compare results between those two samples in this case?

Thanks for your help.

Cheers
Lily

[Michiel Bliemer]

Indeed blocking of efficient designs cannot guarantee attribute level balance within each block. But at least it aims to have some degree of attribute level balance, which is better than simply randomly assigning choice tasks to each block.

The point of blocking is to split the design into smaller parts in order to give respondents a subset of the choice tasks such as to avoid fatigue. Giving 40 choice tasks to a single respondent would be a lot and may lead to fatigue effects and unreliable responses later in the survey compared to respondents that see a smaller number of choice tasks. So while you can compare the results, you may find differences because of fatigue effects.

Michiel