A mosaic plot is a powerful visual tool for displaying the relationship between two or more categorical variables, and learning how to read a mosaic plot enables you to quickly spot patterns, associations, and outliers in contingency tables. Unlike a simple bar chart, a mosaic plot encodes both the marginal frequencies and the conditional proportions within each rectangle, making it especially useful for exploratory data analysis in fields such as sociology, market research, and biology. Think about it: in this guide, we will walk through the components of a mosaic plot, explain the logic behind its construction, and provide a step‑by‑step approach to interpreting it correctly. By the end, you will be able to glance at a mosaic plot and extract meaningful insights without needing to revert to raw counts.
Quick note before moving on.
Introduction to Mosaic Plots
A mosaic plot, also known as a Mekko chart or a Marimekko diagram, represents a contingency table as a series of adjacent tiles. Each tile’s area is proportional to the joint frequency of the corresponding category combination, while its height (or width, depending on orientation) reflects the conditional proportion of one variable given the other. This dual encoding allows viewers to assess both the size of each group and the strength of association at a glance Took long enough..
When you first encounter a mosaic plot, ask yourself three core questions:
- What are the variables being displayed? Identify the categorical axes (usually placed on the x‑ and y‑axes).
- How are the tiles sized? Recognize that larger tiles indicate higher joint frequencies.
- What do the shading or colors convey? Often, color intensity highlights residuals from independence, helping you spot significant deviations.
Answering these questions sets the stage for a deeper reading of the plot.
Steps to Read a Mosaic Plot
Below is a practical, numbered workflow you can follow each time you encounter a mosaic plot. Treat it as a checklist to ensure you do not overlook any visual cue Practical, not theoretical..
1. Identify the Axes and Variables
- Look at the labels on the horizontal and vertical axes.
- Determine which variable is outer (usually the one that creates the columns) and which is inner (the one that creates the rows within each column).
- Note the number of categories for each variable; this tells you how many tiles you will see in each direction.
2. Examine Tile Areas (Joint Frequencies)
- The area of each rectangle = (total sample size) × (joint proportion).
- Larger rectangles mean that the combination of those two categories occurs more often in the data.
- Compare areas across the plot to get a sense of which category pairs dominate the dataset.
3. Assess Tile Heights (Conditional Proportions)
- Within each column (or row, depending on orientation), the height of a tile shows the proportion of the inner variable given the outer variable’s level.
- Here's one way to look at it: if the outer variable is “Education Level” and the inner variable is “Job Satisfaction,” the height of a tile in the “College Graduate” column tells you what fraction of college graduates report each satisfaction level.
- Look for columns where the heights are uneven; this suggests a relationship between the variables.
4. Interpret Shading or Color Coding
- Many mosaic plots use a diverging color scheme (e.g., blue for negative residuals, red for positive residuals) to indicate how much the observed count deviates from what would be expected under independence.
- A tile shaded strongly in one direction means the observed frequency is far from the expected frequency, hinting at a notable association.
- If the plot is unshaded, you can still infer association by comparing the pattern of heights across columns; uniform heights suggest independence, while systematic tilting indicates dependence.
5. Look for Patterns and Anomalies
- Uniform columns: If all tiles in a column have roughly the same height, the inner variable is distributed similarly across that outer category—suggesting little association for that level.
- Systematic tilting: A gradual increase or decrease in tile heights across columns reveals a trend (e.g., higher satisfaction with higher education).
- Outlier tiles: A single tile that is both large and sharply colored may represent a small but important subgroup that deviates strongly from expectations.
- Empty or near‑empty tiles: Very small areas may indicate rare combinations; treat them cautiously because sampling variability can be high.
6. Summarize the Findings in Plain Language
- Translate what you see into a concise statement: “Among respondents with a high school diploma, job satisfaction is evenly spread, whereas college graduates show a clear shift toward ‘very satisfied’.”
- If residuals are shown, add: “The positive red shading in the ‘College Graduate–Very Satisfied’ cell indicates more satisfied graduates than expected if education and satisfaction were independent.”
Following these steps ensures you read the plot systematically rather than getting lost in its visual complexity.
Scientific Explanation Behind the Mosaic Plot
Understanding why a mosaic plot works helps you trust its interpretations. The plot is built from a contingency table (O_{ij}) where (i) indexes the outer variable levels and (j) indexes the inner variable levels. Let (N = \sum_{i,j} O_{ij}) be the total sample size Not complicated — just consistent. Nothing fancy..
-
Marginal proportions:
- Outer variable margin: (p_{i+} = \frac{\sum_j O_{ij}}{N})
- Inner variable margin: (p_{+j} = \frac{\sum_i O_{ij}}{N})
-
Expected counts under independence:
- (E_{ij} = N \times p_{i+} \times p_{+j})
-
Tile dimensions:
- Width of column (i) ∝ (p_{i+}) (the outer margin).
- Height of tile ((i,j)) ∝ (\frac{O_{ij}}{\sum_j O_{ij}} = \frac{O_{ij}}{N p_{i+}}) (the conditional proportion of inner variable given outer level (i)).
- As a result, area ∝ (p_{i+} \times \frac{O_{ij}}{N p_{i+}} = \frac{O_{ij}}{N}), which is exactly the joint proportion.
-
Residuals and shading:
- Pearson residual: (r_{ij} = \frac{O_{ij}
