Have an idea?

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

Coding 'dual-response none' in single CSV format

The CBC/HB manual states "Note: Dual-Response None studies (see Appendix J) cannot be coded using the Single CSV Format." and also "Different respondents could potentially have different numbers of tasks, and different tasks can have different numbers concepts under this layout."

So incorporating the above and treating the none option as an additional task can 'dual-response none' be coded in single CSV format as follows?

For each task code as standard (without the none)
Then replicate the task, with the additional line of zeros for none, with either a zero for choice if none was not chosen (and the original choice indicated as in the 'standard' coding) or a 1 if 'none' were chosen and the original choices zeroed in the choice column.
asked Jun 27, 2012 by anonymous
retagged Sep 13, 2012 by Walter Williams

1 Answer

+4 votes
Is there a reason that you cannot use the dual-CSV format, which supports dual-response none?

Behind the scenes, when we code the tasks for estimation there are three scenarios:

'Would Choose' - The task is written normally with a none option.
'Would Not Choose' - A first task is written without the none, and a second one with the none included and chosen (this is analagous to what you describe).
'Missing (no response available)' - A task is written without the none option.

This can be coded in the single CSV as you suggest, however you will need to handle the none specially as tasks cannot mix none options within the same file.  To do this, you need to add a new column as a 'None' attribute.  Your none concepts will have a 1 for that column (and a 0 in all other attribute columns).  All other concepts will have a 0 in the None column.  In HB, you need to code that column as User-Specified and the project should specify there is no none option.  The new column will react semantically the same as the none option.
answered Jun 27, 2012 by Walter Williams Gold Sawtooth Software, Inc. (15,275 points)
The reason I'm avoiding the dual-csv format is that I'll likely have 1500 versions (is this possible? - most of the limits for the other factors are set at 1000). I'll happily plan to use this if this number is acceptable.

The original specification has att1 (a time of day/week) with 9 levels. There are two further attributes of 15 and 10 levels (both relating to a discount), however, there are many prohibitions and the standard errors are projected to be large. I'm considering reducing att1 to only four for each respondent (three based on their explicit a priori choice and the forth selected to make up the numbers in the least (respondent) selected category). I then intended to generate acceptable values for the levels for att2 and 3 on the fly, relying upon the large amount of respondents and randomization to give an acceptable design, however, this will created a large number of versions.
1500 is unusually large for the number of versions, but there is nothing in the software that should prevent you from using it.  The only possible constraint will be memory, and you will just have to see how it handles it.  Let me know if you have any problems.
Sorry to re open this old thread.
But within the same single task in a single CSV file, when coding "dual response none" (the way that would prevent the loss of choice information), did you mean coding the positive response for "None" as "1" and at the same time coding a positive response for a concept within the same task as "1" if both were selected in the same task?  Basically speaking coding 2 responses as "1"  and "1" in the same task?

If not I am not sure the loss of choice information didn't happen here