Scatterplot (Scatter Diagram)
A scatterplot (scatter diagram or XY diagram) plots pairs of numbers as points on two axes to show how two variables move together. It reveals the direction, form and strength of an association and flags outliers or clusters. It is one of the seven basic quality tools used to investigate possible causes of a problem in Lean Six Sigma.
A scatterplot is also referred to as a scatter diagram and an X-Y diagram: it plots pairs of data points on a two-dimensional chart. It is used to determine the strength and direction of association between two variables and is one of the seven basic quality tools used in process improvement. Unlike a histogram, which shows one variable’s distribution, a scatterplot compares two variables at once. In a Six Sigma project you normally put a suspected cause (X) on the horizontal axis and the outcome (Y) on the vertical axis, following Y = f(X).
How do you read a scatterplot?
Read four things before calculating a correlation coefficient. This quick visual check stops a misleading number from driving the conclusion.
| Look for | What it looks like | What it suggests |
|---|---|---|
| Direction | Points rise left to right (positive), fall left to right (negative) or show no slope | Whether Y tends to increase or decrease as X increases |
| Form | A straight band, a curve, a U shape or no pattern | Whether a straight-line model is suitable; a strong curve can have a correlation near zero |
| Strength | A tight band versus a broad cloud | How predictable Y is from X over the range observed |
| Unusual points | Isolated points, gaps or separate clusters | A recording error, a special cause, different process streams or an influential extreme value |
Also ask whether the axes and units make sense. A truncated axis can make a weak pattern look dramatic. If points form two clusters, label them by machine, shift or supplier before fitting a single line; the combined pattern may hide the story within each group.
How do you make a scatterplot?
- Write the question. Choose one Y and one suspected X, such as “Does cure temperature affect peel strength?” Define how each will be measured and confirm the measurement system is fit for purpose.
- Collect paired data. Each row must be the X and Y from the same unit, run or time. Do not sort the two columns independently. Keep the date, machine and other likely grouping fields.
- Put X on the horizontal axis and Y on the vertical axis. Use sensible, evenly spaced numerical scales and label the units.
- Plot one point per pair. If many readings overlap, use transparency, smaller markers or a count/hexbin view rather than hiding the duplicates.
- Read the pattern. Check direction, form, strength and unusual points. Investigate an outlier before excluding it.
- Quantify and follow up. If the pattern is roughly linear, compute Pearson’s r and fit a regression line; inspect residuals. Use a designed experiment when you need evidence of cause and effect.
In Excel, put X in the first column and Y in the next, select both, then choose Insert → Scatter (X, Y) → Scatter with only markers. Add axis titles. If appropriate, add a linear trendline and display its equation and R-squared. Do not choose a Line chart: it treats the horizontal labels as equally spaced categories rather than numerical X values.
Worked example: cure temperature and peel strength
The ten paired readings below are illustrative. They were chosen to show the method, not taken from a real plant. A team suspects that the cure temperature X (°C) affects peel strength Y (N).
| Run | Temperature X (°C) | Peel strength Y (N) |
|---|---|---|
| 1 | 150 | 21.0 |
| 2 | 155 | 22.4 |
| 3 | 160 | 22.1 |
| 4 | 165 | 24.0 |
| 5 | 170 | 24.6 |
| 6 | 175 | 25.9 |
| 7 | 180 | 25.1 |
| 8 | 185 | 27.3 |
| 9 | 190 | 27.8 |
| 10 | 195 | 29.0 |
The plot above shows a strong, roughly straight, positive relationship. The Pearson correlation is r ≈ 0.980, and the least-squares line is Y ≈ −4.436 + 0.1702 X, with R² ≈ 0.960. Over the observed range, a 1°C rise in X is associated with about a 0.17 N rise in predicted strength. The fitted intercept is only part of the line equation; it is not a prediction of physical strength at 0°C, far outside the data.
This is promising evidence for an investigation, not a license to turn up the oven. All ten readings might also have been taken in time order while a material lot changed or a machine warmed up. Randomizing temperature in a controlled experiment would help separate the effect of temperature from those other influences. See the correlation and regression guide for the formulas, residual checks and limits.
Scatterplot versus other quality charts
- A histogram shows the distribution of one numerical variable, such as the spread of peel strength.
- A Pareto chart orders categories by count, such as defect types.
- A run or control chart plots a measure in time order, showing shifts or instability that a scatterplot can hide.
- A scatterplot compares two numerical variables, looking for an association that could explain a problem.
The tools work together. Start with a check sheet to collect paired data, use a time plot to check stability, then use the scatterplot to explore candidate causes. In DMAIC this is most often Analyze work, after the data are defined and measured in Measure.
Common mistakes
Correlation is not causation. A third variable can move both X and Y, time trends can create a false relationship, and selection bias can change the apparent pattern. A single outlier may also create or erase a high r. Do not extrapolate beyond the observed range, do not delete points just to improve the fit, and do not rely on r alone: it measures only linear association. Always look at the plot and ask what process story could have produced it.
Sources and further reading
Method reference: ASQ: Scatter Diagram.
Frequently asked questions
What is a scatterplot used for in Six Sigma?
A scatterplot is used to investigate whether a suspected input X is associated with an output Y. It shows the direction, form and strength of the relationship and flags outliers or separate process streams.
What are the three main types of correlation on a scatterplot?
Positive correlation rises from left to right, negative correlation falls, and no apparent correlation shows no clear pattern. Curved relationships and clusters are also possible, so do not stop at those three labels.
Does a scatterplot prove cause and effect?
No. It shows association, not proof of causation. A third variable, a time trend or an outlier may explain the pattern; a properly designed experiment provides stronger causal evidence.
What goes on the X and Y axes?
Put the suspected cause or input on the horizontal X axis and the effect or output on the vertical Y axis. Each plotted point must be a matched pair from the same unit, run or time.
See also: correlation and regression, design of experiments, the seven quality tools, and control charts.
The other basic quality tools in this guide: Pareto charts and fishbone (cause-and-effect) diagrams.
Related guides
Last updated: · Published by Lean Six Sigma Training Guide (sixsigmatrainingfree.com) · How this guide is written