Stratification

Stratification, also known as flow or run chart, is one of the 7 Basic Tools of Quality in data analysis. It is relevant to the field of manufacturing as it allows business leaders to identify patterns within multiple data sets.

What Is Stratification?

Stratification is a mode of data analysis where data is grouped into homogenous groups – called strata – for visual graphical representation.

Each stratum is from a different data source, and is represented differently on the stratification chart according to a visual legend.

The purpose of a stratification chart is to allow someone to see patterns between and different sets of homogenous data. It is not as in-depth, statistically or mathematically-speaking, as a control chart.

Here is an example of a stratification graph:

Stratification vs. Control Chart: What’s the Difference?

Stratification and control charts are both basic tools of quality control, but there are a couple significant differences:

• Control charts always show a control limit, whether a ceiling or floor or both
• Stratification charts are line graphs with no identified control limits.

A control chart should be used when attempting to see if a process is stable, or with control limits. A stratification chart should be used when investigating the possibility of trends across multiple data sources.

Note that this is also different from a check sheet, one of the other 7 tools of quality.

Different Types of Stratification

Earlier, we referred to stratification by other names: flow charts or run charts. For our purposes in explaining basic core concepts, these are all the same thing. However, there is a caveat that there are critical differences between types of stratification analyses.

For example, a run chart is a subtype of stratification chart that includes a time sequence. This is often shown along the X-axis.

When using a run chart in particular, the focus in this style of data analysis is measuring a variable against increments of time. An example of this would be measuring units produced on a single production line over the course of a day.

The Manufacturing Applications of Stratification

A common use of stratification is in organizing survey data. In this example, data points are visually separated by color or by drawing trend lines.

Because stratification is used for detecting patterns, they can include all sorts of data. Here are some examples of data sources that can be used, even within the same chart:

• Shift changes
• Workforce turnover
• Machine performance/OEE
• Machine downtime
• Defects
• Products
• Materials or sub-components
• Supply chains

You may have noticed that this list is incredibly broad. This is because stratification is a very applicable tool by design.

Putting stratification into its simplest terms, it is a way to visually see patterns in data. This means that you can use pretty much any type of data you’d like, as long as you set it up properly by ensuring each stratum is fairly represented according to your specifications.

For example, a shop floor manager would be able to see several groups or “buckets” of data points at once, such as number of workers present each day, the number of defects produced per shift, and number of final products shipped.

These 3 separate strata – workers, defects, and final products – may have statistically significant patterns between them or within each of them, depending on how the data shows up visually.

Interpreting Stratification Charts

It’s important to choose the right application for your stratification chart, and that includes following some basic guidelines.

Necessary Things to Include In Stratification Charts

Stratification charts are quite simple analysis tools, but there are definitely elements that must be included. These are:

• Legend: It’s important to include a key or legend to your chart so that viewers can understand the differences in color or shape for various data points.
• Homogenous units per stratum: Your selection of types of data (as separated within the legend) may be measured using different metrics, but all data points within one stratum must be measured consistently. For example, if one stratum is time spent assembling a component, you cannot have points of data referring to minutes in one place and hours in the next.

Correlation vs. Causation

It is important to remember that stratification charts only show correlation, not necessarily causation. For example, just because similar patterns between strata may emerge within your stratification chart, doesn’t mean that one action causes the other. They may in fact just be correlative. This is why it is helpful to combine stratification with one or many other tools of quality. Many perspectives of the same data sets can expose insights and data collection biases alike.