A histogram is a bar graph that summarizes the frequency of occurrence of information that has been collected over time. They visually display summarized data showing the frequency, shape, and central tendency of the information. The first action I take when dealing with variable or continuous data is to construct a histogram so that I can see how the data looks. It helps me answer questions such as, “What does the distribution look like?” “Is it normal?” “Is it shifted from the target value?” And so on. Of course, you can easily construct a histogram using any statistical software, such as Minitab or even Excel.
A histogram can be used to:
- Pictorially summarize large amounts of data
- Visually see process centering, spread, and shape
- Provide useful information for predicting future performance of a process (if the process is stable)
- Observe a change in the process by comparing two histograms
- Observe patterns in the data
- Provides clues to reducing variation and causes of problems
- Observe the production consistency of a quality characteristic
- Graphically show the relationship between the capability of the process and the engineering specifications
To construct a histogram:
- Collect the data. Typically, 30 or more measurements are required.
- Decide on the classes or intervals for grouping the data
- Create a tally sheet using these classes
- Complete the tally sheet and summarize the data
- Draw the histogram so that the height of the rectangle represents the number in the tally for that class
An example of a histogram, with the lower and upper specification limits added, is shown below. Since the distribution is using the whole tolerance (USL – LSL), a useful strategy to improve this process would be to identify the various variation categories that are contributing to excess variation and look for ways to reduce the variation in each category. I typically start using the 6Ms, i.e., Man, Machine, Method, Material, Measurement, and Mother Nature. The goal would be to continually reduce the width of the distribution by removing variation.
A famous statistician, Ellis Ott, is quoted as saying, “Always, always, plot your data!” I have found that this is an excellent habit to get into.