4 Replies Latest reply on Jan 29, 2017 1:02 PM by Shawn Wallwork

    Building lift/ROC charts for model comparison

    Suneil Shrivastav

      Hello! This is my first time on the Tableau community, and I come bearing a question

       

      I have an underlying Teradata customer risk profile model (output is a binary indicator) which scores data once a month. I need to plot this curve where on the X axis I have the % of population in 10% increments (deciles), and on the Y-axis the % of all people who have been modeled.

       

      There need to be two lines - the gain curve (something like a Pareto curve, but with ten inputs on X and 10 on Y) and a random (0,0) to (1,1) line on the same chart. This second line shows the % of people modeled accurately if it were done at random, and basically would be a 45 degree line from bottom left to top right.

       

      The other line is the model line.

       

      My problem is this - since the model has numbers starting from the first decile only, and no rows exist where the decile is 0, these lines do not connect on the origin as they should in a traditional gain chart (for instance Understand Gain and Lift Charts - Listen Data )

       

      Is there a way to get the curves to connect at (0, 0)? I can imagine there should be some way to create a dummy variable that has 11 instantiations for the ten in the original variable, but I don't know how to do it