2 Replies Latest reply on Oct 9, 2012 9:14 AM by james.hayes

# Correlations of Dimensions - how to compare dimensions when that's all you have?

I have a table full of text, names, Yes/No values for various pieces of equipment and no numbers anywhere.  The goal is to find high correlations between the values in the dimensions.  One key dimension holds a value if the electric switch has failed in this particular piece: it's value is either "yes" or "no".  Based on this value, all the failed switches can be researched against all the other values in the row to find which values a failed switch is most correlated with.  Is it the value of "v120 power supply" in the Volt field?  Or perhaps its "2 year warranty" in the Notes field.

• ###### 1. Re: Correlations of Dimensions - how to compare dimensions when that's all you have?

Hi,

You can turn dimensions into usable measures with a calc like SUM(IF [switch failed]="yes" THEN 1 ELSE 0 END). Then you can do % of Total with a calc like [No. of Failed Switches]/SUM([Number of Records]), or [No. of Failed Switches]/TOTAL(SUM([Number of Records])).

For the correlation work, there are a various ways to do this. A simple way to compare against a single dimension would be to create a bar chart with the No. of Failed Switches measure on Columns, the dimension of interest on Rows, and then use the one-click sort to find the most common reasons. You could add additional dimensions to the view to see where the most failures occur within all the charts created by a small multiple view.

Jonathan

• ###### 2. Re: Correlations of Dimensions - how to compare dimensions when that's all you have?

Jonathan, thanks for the reply.  I would need to know all the various values the dimensions could be if I were to be able to hard code each one and assign it a value.  I was hoping to avoid manually assigning numerical values to alphabetic data.

Your simple way to find a correlation is interesting, and I think it brings to light a new feature that could make Tableau better: the ability to find correlations automatically.  Based on Rows, such an algorithm could compare every value, whether numeric or alphanumeric, against every other value in the database (or portion thereof).  Matching patterns of values creates a higher correlation.  These correlations could be presented to the user in report form for review.  The best ones, could be selected as "the" correlation(s) for further display charting or dashboarding options.