A heatmap shows patterns, trends, and correlations in data. It does this by using color as a direct representation of the values. By adding a date or a time scale on the x-axis it shows how the values evolve over time.
A heatmap can serve as a perfect starting point to get the lay of the land. It lets you explore the data and it gives hints on where to look for other outliers, other viewpoints, or specific angles. Combining a heatmap with other charts might be a good idea if your goal is to tell a more detailed story.
A heatmap can have the shape of a table or a matrix or it can function as a color layer over a geographical map. If shaped as a matrix, a heatmap is a perfect way to reveal correlations. A heatmap as an actual map shows the density of a value at a certain place or area.
Heatmaps can also be seen as a layer over other chart types. Coloring those charts numerically uses the same visualization logic and has the same impact as a ‘regular’ heatmap.
If you want to zoom in on one category and focus on the evolution of that variable, you can use heatmaps in only one dimension. These charts are very popular in climate communication and often visualize temperatures.
If there is a geographical dimension to your data, you can add a color layer to a map and show the density of a value at a certain place or area.
Although a choropleth map is visually similar to a geographical heatmap, it does show data in a different way. In choropleth maps, colored regions correlate with geographic or artificial boundaries. The color shows a proportional value, such as an average, for one of those delineated regions
When both dimensions in the data are categorical, we can replace a heatmap with parallel coordinates. Instead of using color to represent value, the parallel coordinate uses the location of the categories on the axis to denote the values.
A heatmap data with an evolution dimension is easily transferable to a multi-series line chart. The correlation factor will be lost a bit but instead, you will be able to compare the values between the series much better.
If you want to keep showing correlations in the data, and you want to plot the numerical values more specifically on an axis instead of in bins, you can choose a scatter plot. Binding the marks to the color or the size shows the density of a value.
If your data has an order to it, meaning that it is somehow sortable, a numerical scale is the one to go with. If the data is nominal, you should choose a categorical one.
When the data only varies in one direction, a sequential scale is the best choice. When your numerical data has a logical breakpoint and the data varies in two directions, a diverging scale is the way to go.
Finally, there is a difference between stepped or continuous scales. With data that is not continuous, but ordinal, you should always go for a stepped scale.
But with continuous data, we can choose what scale we want. Choosing a stepped scale for continuous data helps you make your point more clear and lets your readers derive values more easily. A continuous scale gives a more nuanced view and allows more interpretation up to the reader.
Sorting the columns in a tabular heatmap is not always possible.
If your x-axis is numerical or temporal, you cannot sort it at all.
If it is categorical, and there is no order to be followed, sorting it ascending or descending might improve readability. This also goes for the categorical Y-axis.
If you do want to add an extra layer of detail to your heatmap, you can add data labels to every ‘cell’ in the matrix. This also works the other way around. If you have a flat table or data sheet, adding a color layer to the values can instantly help the readability and comprehension of the data.