Categorical Data Analysis: Sometimes That’s All You Can Get and That is Okay

      If you work in manufacturing, data collection is not usually a problem.  By their nature these operations give off numerous amounts of data and information.  The problem is usually to capture all of it and then filter out what you need from the whole vast collection.

      The world of services and transactional six sigma is more of a challenge.  By their very nature, these operations do not throw off the vast quantities of continuous data.  Yes, transaction counts and time duration are sometimes available. But many transactional process vary significantly in complexity within an area of study so you cannot make the assumptions we do in manufacturing.  It’s not thousands of one uniform product rolling down a conveyor belt.

      What is one to do?  Frequently transactional data is classified into categories.  For example, customer service operations classify the resolution of a customer contact into categories and then various subcategories.  On numerous occasions, teams I work with use Pareto analysis to drill down to root cause of problem areas of focus in these types of operations.

      This is definitely not an ideal approach to take.  But sometimes that is all the data you have and there might be a lot of it to work with.  Also many operations have cyclicality to their data so the frequently espoused idea to ‘take two or three months’ worth of data and use that’ could be a bad idea.  There’s can also be a ‘Hawthorne Effect’ too.  Both of these issues are not generally brought up or acknowledged by the ‘data experts’ who should know better.

      So the next time you have a process improvement opportunity but only categorical data, don’t lose heart.  You can use six sigma rigor to perform the analysis and still improve a process.  The results may positively surprise you.