Scatter Diagram: How to do it
The Quality Toolbook >
Scatter Diagram > How to do it
When to use it | How to understand it |
Example | How to do it | Practical
<-- Previous |
How to do it
- Determine the two items that you wish to compare. One may be identified as the suspected cause and the other as the suspected effect. This may come from the use of other tools, such as a Cause-Effect Diagram or Relations Diagram.
- Identify the measurements to be taken. Both must be variables (i.e. measurable on a continuous scale) and it must be possible to measure both at the same time.
Make the measurements as specific as possible in order to reduce variation and increase the chance of a higher correlation. For example, measurements from a single supplier's materials may be better than measuring all supplied materials.
- Make 50 to 100 pairs of measurements. When doing this, aim to keep all other variables as steady as possible, as they could interfere with the final figures.
Be very careful when measuring human behavior, as the very act of measurement can cause the measured people to change their behavior, especially if they suspect they may lose out in some way.
- Plot the measured pairs on the Scatter Diagram. Design the axes and scales on the diagram to give the maximum visual spread of points. This may involve using different scales and making the axes cross at non-zero values (as in the figure below).
If investigating a possible cause-effect relationship, plot the suspected cause on the x-axis (horizontal) and the suspected effect on the y-axis (vertical).
Fig. 1. Setting the scale
- If the correlation is high, a regression ('average') line may be drawn
through the plotted points, to emphasize the trend. This may be
calculated or estimated by eye (although this should be made clear to any future readers of the diagram).
- If the correlation is reasonably linear, then a correlation coefficient may be
- Interpret the diagram and act accordingly. This may be to identify improvements or to enable estimation of future effect values. If it is the latter, the standard error may be calculated, as in the figure below.
When using a Scatter Diagram to estimate future effect values, only estimate within the known range of correlation, as the shape may change outside that range.
Note: Scatter Diagram calculations are on a
<-- Previous |