Have an idea?

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

How can I merge raw MaxDiff data across markets before HB analysis?

Dear all,

I am planning to run 5 separated MaxDiff exercises via the SSI Web 8 version for 5 different markets.

However, since my number of respondents is somewhat limited within the markets (min = 30, max = 65), I would like to combine/merge the raw data before I run the HB analysis.

Is there a way to export the five raw data sets from SSI Web 8 and then combine/merge them manually before uploading the combined raw data set again to SSI and running the HB analysis on the combined data?

If so, any help and explanations on how to do so is greatly appreciated!

Thanks in advance and best regards,

asked Aug 1, 2018 by JoGu (250 points)

1 Answer

+1 vote
There is probably more than one way to do this, but my approach would be:

1.  Export each country's MaxDiff datafile to a .CHO file (this is a space-delimited text file that you can open in Word or Wordpad if you wanted to.

2.  Using Word or Wordpad, combine the separate .CHO files into a single .CHO file.  Keep track of which respondents are from which country in the sequence of the file.

3.  SSI Web cannot read .CHO files (it can only write them), so you'll need to use the standalone CBC/HB estimation software to estimate the HB scores on the combined .CHO file.  If you have a CBC license within SSI Web, then you automatically have the right license to install and run CBC/HB standalone software.  Download and install CBC/HB standalone software from: http://www.sawtoothsoftware.com/support/downloads/download-cbc-hb

4.  After you launch the CBC/HB software, open a new project and point it at the .CHO file, and run the utility estimation, then you need to follow the instructions in the CBC/HB Help Manual, Appendix K "Estimation for MaxDiff Experiments" to do a little bit of rescaling manipulation in Excel to put the scores on the 0-100 probability scale.

[Note: when you estimate scores in CBC/HB, it will look like the last item is "missing" from the score estimation.  It is really not missing, since CBC/HB software sets the utility of the last item to 0 and scales the remaining items with respect to this last item assumed to be 0.  So, Appendix K will explain to you that you need to insert that last item's reference level of 0 back into the set of scores prior to doing the subsequent rescaling step to put them all on a 0-100 probability scale.]
answered Aug 3, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (160,485 points)
Thank you very much for your answer!

I downloaded the CBC/HB standalone software and followed your instructions. Now some more questions came up:

1. As far as I understand, beside the .cho file with the merged data I also need to upload a .att, a .eff, a .mtrx, a .position and a .val file into the project to make a proper estimation. SSI Web exports these files automatically when exporting the .cho file. I opened the files with wordpad but I am not sure if I completely understood what they are used for. Can I just copy the files of one of the surveys from which I merge the data and rename them according to the file with the merged data?

2. After the estimation in the  CBC/HB standalone software some .csv files are created, one of them named xxx_utilities.csv, which I canĀ“t read intuitively. When conducting the HB estimation directly within SSI Web (for a single data set, whitout merged data) a nice .xlsx file named MaxDiff_scores_hb_report.xlsx is created. It displays the MaxDiff utilities in a well-arranged and easy to read way. Is it also possible to get such a report from the CBC/HB standalone software?

Best regards!
If you only use a .CHO file and there are no other optional files, then default labels and default settings will be used.  Most of the time that's what I do.  But, you'll probably want to use one of the .ATT files (just need one to apply to all of your merged respondents...give it the prefix name matching the merged .CHO file) which provides the labels for the items.  

.EFF is unnecessary for this analysis run.  .MTRX is not needed, since CBC/HB software will recognize upon reading in the .CHO file that his is MaxDiff data and what kind of prior covariance matrix to apply.  .val and .position files are not needed.

The .csv file that is produced by CBC/HB should be readable for you within a program like Excel (you'll find scores for your K-1 items, where the last item you need to fill in as a raw utility of zero).  You'll need to open it and follow all the instructions for rescaling the scores within that Excel spreadsheet as contained in Appendix K of the CBC/HB manual.
Thank you very much for the clarification about the file  types.
I also managed to open the .csv file in a readable way. I had to adapt the number format settings of my computer. The rescaling worked as well.