Stratification: It’s a matter of perspective

During a lean project, you may find yourself with a huge collection of data points. Although this represents a great resource of information pertaining to the process, the numbers themselves do not tell a story. Every data set requires interpretation and it is this perspective that provides context to the numbers. Think of key stratification factors when trying to understand a process.

The purpose of stratifying data is to break up the initial data set into smaller subgroups in order to determine significant causes or trends. This is important as numbers may deceive and lead you astray if you react solely on intuition or “gut” instinct.

Consider the following:
A manager is presented with the below process graph regarding the output yield for a particular manufacturing line in calendar year 2014.

The manufacturing line has an average yield of 82.4% for the calendar year 2014. There are no significant trends that jump out immediately, however, there are a cluster of points above the average from April to July which may suggest a good process.

The manufacturing line produces two finished products, therefore, the data was stratified into Product A and Product B.

Product A was run in Q1 and Q3 of 2014 with an average yield of 75.3%. The range is 61 – 97%.

Product B was run in Q2 and Q4 of 2014 with an average yield of 89.3%. The range is 74 – 100%.

It was also noted that the manufacturing line is staffed with two crews, therefore, the data was stratified into Crew 1 and Crew 2.

Crew 1
Produced Product A with average yield of 84%, range: 72 – 97%.
Produced Product B with average yield of 93%, range: 84 – 100%.

Crew 2
Produced Product A with average yield of 68%, range: 61 – 75%.
Produced Product B with average yield of 85%, range: 74 – 95%.

From these few graphs, we can observe that the process for Product B results in higher yields than Product A. Furthermore, Crew 1 is getting a significantly higher yield in comparison to Crew 2 for both Products.

With each new stratification factor, the story for the process evolves. You can therefore appreciate that the process improvement actions would differ depending on the trends uncovered with each new stratification analysis performed.

Further stratification analysis could be broken down as follows:

• What are the yield percentages for product changeovers compared to campaign run?

• What are the yield percentages for the crews based on headcount? (i.e. when there are members on vacation or sick?)

• What are the yield percentages based on process deviations? (i.e. is there a link between process deviation occurrence and yield?)

• What are the yield percentages based on equipment run parameters?

Of course, further statistical analysis can be performed for each new subset of figures to validate the correlation between factors.