2 Replies Latest reply on Apr 26, 2017 7:23 AM by Jamieson Christian

    FINALLY! An elegant solution to the "null vs. zero" issue

    Susan Baier

      As many of you working with survey software like Qualtryx and SurveyGizmo are probably aware, these tools don't distinguish between someone failing to select one "check all that apply" item (but who selected others) and someone simply skipping the question entirely.

       

      Steve Wexler at Data Revelations has described in the past how critical this differentiation is when visualizing CATA questions (and he's been helping me struggle through how to resolve it). And now, thanks to Steve and the mighty Joe Mako , we can easily overcome this issue WITHIN TABLEAU -- which makes it the perfect solution for those of us who don't use data prep software like Alteryx (or don't want to add another lengthy component to their Alteryx workflow) or who want to take advantage of the new Qualtrics Web Data Connector.

       

      Read the solution in a new post on Steve's blog at Data Revelations.

       

      Thanks as always Steve for your support of the survey data users among us! Look forward to seeing both you and Joe in Austin for TCC16!

        • 1. Re: FINALLY! An elegant solution to the "null vs. zero" issue
          Alissa Swartz

          In Qualtrics, you can mark the seen but unanswered data as -99. Does that solve your problem/?

          • 2. Re: FINALLY! An elegant solution to the "null vs. zero" issue
            Jamieson Christian

            Alissa Swartz : That was the workaround that Qualtrics recommended as well, to deal with the fact that their Web Data Connector is incapable of putting nulls into the output. But there are three problems:

             

            1. The option to substitute non-responses with -99 is only available for file exports. There are no configuration options for the Web Data Connector.
            2. The alternative, then, is to use custom response codes to ensure that zero is never a valid (internal) data value. At that point, you can interpret it as a non-response. The problem is, this method only works for responses going forward. If you want to get an existing survey working with the WDC, this option won't help for historical responses.
            3. -99, and even zero, adversely affects any aggregations. So you pretty much have to create a calculated field for every single numeric response to convert the non-response value to null. It's doable, but it makes a huge mess of your data source. (Pivoting, as described in the Steve's blog article, mitigates this mess substantially — but that's only available for files, not WDC data sources.)

             

            Susan Baier : It's important to understand that Qualtrics, at least, does know the difference between a "zero response" and a "non-response". The issue is simply that their Web Data Connector does not properly interact with the SDK to generate nulls where nulls are needed. We have asked them to deal with this bug, and presumably they have ticket open internally, but I have yet to get traceability or a commitment to fix the bug in a specific timeframe.

             

            And with that, I'll stop resurrecting old threads.