Advanced Graph Creation

Graph Types such as Bar Chart, Histogram, and Scatter Plot draw graphs in a fixed format with minimal setup. Use Custom Graph when the graph you want does not fit a fixed format: overlaying a regression line on a scatter plot, overlaying a density curve on a histogram, or comparing subgroups with facets.

Custom Graph is based on the Grammar of Graphics, a framework that treats a graph as a combination of components. Once you understand the basic components, you can create diverse visualization patterns. This page explains each component by building graphs from the Auto MPG dataset (fuel efficiency data for 398 cars from 1970-1982).

Components of Grammar of Graphics

A Custom Graph consists of one or more Layers and graph-wide settings. A layer defines a single drawing and holds the following elements.

Aesthetics: The mapping between variables (dataset columns) and visual attributes (position, color, size, etc.)
Geometry: The visual representation of data (Point, Line, Bar, etc.)
Statistics: Statistical transformations of data (binning, smoothing, etc.)
Position: Position adjustment of elements (stacking, dodging, etc.)

The following elements are specified for the whole graph and shared by all layers.

Data: The dataset to visualize
Coordinates: Coordinate system (Cartesian coordinates, axis swapping, etc.)
Facets: The structure when creating graphs composed of multiple small graphs
Scales: How to convert data values to visual values

Overlaying multiple layers combines several drawings, such as a scatter plot and a regression line, into a single graph.

Aesthetics - Mapping Visual Attributes

Map data columns to visual attributes. The most basic is mapping two continuous variables to the x and y axes.

Data: Auto MPG
Aesthetics: x = weight, y = mpg
Geometry: Point

Custom Graph basic scatter plot: visualizing the relationship between weight and mpg (fuel efficiency) in the Auto MPG dataset. Heavier cars tend to have worse fuel efficiency

A negative association is visible: heavier cars have worse fuel efficiency. The Geometry in this example is Point, which draws each data row as one point. Changing the Geometry draws the same mapping in a different form, such as lines or bars.

Mapping to Color

You can add more information by mapping a third variable to color. For example, let's color-code by origin (usa, europe, japan).

Aesthetics: x = weight, y = mpg, color = origin

Custom Graph scatter plot with color coding by origin: displaying usa, europe, and japan in different colors to visualize regional characteristics

Mapping to Size

Map horsepower to point size.

Aesthetics: x = weight, y = mpg, color = origin, size = horsepower

Custom Graph scatter plot combining color and size: color-coded by origin with point size varying by horsepower. Expresses 4 variables in one graph

Larger points indicate higher horsepower. You can visually understand the relationship: heavy, high-horsepower cars have poor fuel efficiency.

fill and color

Aesthetics has two color specifications: fill and color. fill applies to the interior of geometries such as Bar and Area. color applies to outlines, strokes, and points such as Point markers and Line strokes. Point and Line do not support fill; their color is controlled entirely through color. Conversely, Bar does not support color; bar color is controlled through fill.

Layers - Overlaying Layers

Layers allow you to overlay multiple graphs. For example, let's add a LOESS (locally weighted regression) smoothing curve on top of a scatter plot.

Layer 1: Point (x = weight, y = mpg)
Layer 2: Line + Smooth statistic (method = loess)

Custom Graph layer feature: LOESS smoothing curve overlaid on scatter plot. Layers combine multiple graph types into one graph

The blue line is the LOESS smoothing curve. It summarizes the local pattern of the data as a smooth curve. Change Method to Linear (LM) to draw a linear regression line instead. When LOESS is selected, Span adjusts the smoothness of the curve. When points are grouped by aesthetics such as color or fill, the Smooth statistic fits a separate regression for each group.

Turning on Compute interval band overlays a regression interval as a shaded band. Interval type selects between the prediction interval, which covers a newly observed single value, and the confidence interval, which reflects the precision of the fitted curve; set the level with Interval level. See Prediction interval in the glossary for its definition and assumptions, and the Custom Graph Reference for the full list of Smooth parameters.

Statistics - Statistical Transformations

You can display data not just as-is, but after statistical transformation.

Histogram (Binning)

To see the distribution of fuel efficiency, we divide the data into bins (intervals) and count them.

Aesthetics: x = mpg
Geometry: Bar
Statistics: Bin (bins = 20)

Custom Graph histogram with Bin statistical transformation: dividing mpg (fuel efficiency) data into 20 bins and counting. Visualizes data distribution and skewness

You can see that most cars are concentrated in the 15-30 mpg range. The distribution is right-skewed, with fuel-efficient cars being a minority.

Adjust the number of bins based on the shape of the distribution.

Density Estimation

Instead of bins, you can express the distribution as a smooth density curve.

Aesthetics: x = mpg
Geometry: Line
Statistics: Density

Custom Graph Density statistical transformation: expressing mpg distribution as smooth density curve

This expresses the same data distribution in a different way. The density is estimated with a Gaussian kernel, and the bandwidth is chosen automatically by Silverman's rule of thumb.

Comparing Densities Across Multiple Groups

By overlaying density curves for each category, you can compare distribution differences. Let's draw density curves on top of histograms for each origin.

Layer 1 (Bar):
  Aesthetics: x = mpg, fill = origin
  Geometry: Bar
  Statistics: Bin (bins = 30)

Layer 2 (Line):
  Aesthetics: x = mpg, color = origin
  Geometry: Line
  Statistics: Density (Y Scale = Count)

Custom Graph utilizing multiple layers: histogram color-coded by origin (Layer 1) with density curves (Layer 2) overlaid. Compares distribution differences between groups

Layer 1 uses fill = origin to split the bar fill, and Layer 2 uses color = origin to split the density curve strokes. The same color scale is applied to both layers, so bars and curves for the same category share the same color.

Y Scale = Count in Layer 2 multiplies the density by the sample size and bin width to produce a count-scaled curve. This bin width is computed internally with Sturges' formula, independent of the bins = 30 in Layer 1, so the two y-axes are not guaranteed to match exactly.

It's clear that Japanese cars peak on the high fuel efficiency side, while US cars peak on the low fuel efficiency side.

Position - Position Adjustment

Position adjustment becomes important when comparing multiple categories in bar charts.

Stacked Bar Chart

Aesthetics: x = model_year, fill = cylinders
Geometry: Bar
Statistics: Count
Position: Stack

Custom Graph Position adjustment: stacked bar chart using Stack. Shows breakdown of cylinders by model_year as stacked bars

The breakdown of car types for each year is shown as stacked bars. 8-cylinder cars were common in the early 1970s, with 4-cylinder cars increasing toward the end.

Grouped Bar Chart

Changing Position to dodge displays them side by side.

Position: Dodge

Custom Graph Position adjustment: grouped bar chart using Dodge. Bars for each cylinder count placed side by side for easy comparison of trends

This makes it easier to compare trends for each cylinder count. Use Stack to see overall composition, and Dodge to directly compare values between groups.

Stack as 100% Proportions

Setting Position to Fill normalizes each stacked bar so its total length equals 100%. Each segment's length shows its share of the total at that x value. When the data contains negative values, the sum of absolute values counts as 100%, with positive segments stacked above the baseline and negative segments below it. Useful when you want to compare category proportions over time or across groups.

Position: Fill

Reduce Point Overlap

Setting Position to Jitter adds small random displacement to each point so overlapping points become distinguishable. When x is numeric, points are displaced in both the X and Y directions; when x is categorical, only in the Y direction. It helps reveal density in Point scatter plots where many observations share the same x value. Unlike Dodge, which shifts along category boundaries, Jitter uses random displacement, so individual point positions vary slightly each time the plot is redrawn.

Coordinates - Coordinate System

Flipping Axes

Flipping histograms and bar charts horizontally makes long labels easier to read.

Aesthetics: x = mpg
Geometry: Bar
Statistics: Bin
Coordinates: Flipped

Custom Graph Coordinates adjustment: histogram using Flip (axis swap). Swaps vertical and horizontal axes, optimal for displaying long labels or utilizing vertical space

The vertical and horizontal axes are swapped, displaying the histogram horizontally. Useful when category names are long or when you want to effectively use vertical space.

Splitting and arranging graphs by category makes subgroup comparison easier. The Facets section has two types: Facet Wrap (division by single variable) and Facet Grid (matrix division by two variables).

Facet Wrap divides data by one variable and arranges multiple panels in a grid. Any column in the dataset can be used for division. Columns whose number of unique values exceeds the limit cannot be selected. The limits default to 20 categories per variable and 50 panels in total, and both can be changed in Settings.

Aesthetics: x = weight, y = mpg
Geometry: Point
Facets: Type = Facet Wrap (Single Variable)
  Variable = cylinders

Custom Graph Facet Wrap feature: scatter plots divided and arranged by cylinders. Panel division by single variable makes subgroup comparison easy

You can compare the weight-fuel efficiency relationship side by side for 4-cylinder, 6-cylinder, and 8-cylinder cars. 8-cylinder cars are generally heavier and concentrated in the poor fuel efficiency range.

As another example, you can also divide by origin:

Aesthetics: x = weight, y = mpg
Geometry: Point
Facets: Type = Facet Wrap (Single Variable)
  Variable = origin

Custom Graph Facet feature divided by origin: three panels for europe, japan, and usa arranged in a 2×2 grid to compare weight-fuel efficiency relationships by region

The data is split into one panel per origin (europe, japan, usa), and the panels are placed automatically in a grid as close to square as possible. Three panels become 2 columns × 2 rows. To arrange them in a single row, set Columns to 3.

Facet Wrap has options to control panel arrangement:

Variable: Variable to use for division
Columns: Number of panels per row
Rows: Number of rows
Scales: How axis scales are shared across panels

Fixed (the default) under Scales uses the same axis range for all panels; Free X / Free Y / Free give each panel its own axis range. If only Columns is specified, row count is calculated automatically. If only Rows is specified, column count is calculated automatically. If both are omitted, optimal arrangement is calculated based on panel count.

Facet Grid arranges panels along rows and columns using one or two variables.

Facets: Type = Facet Grid (Two Variables)
  Rows = origin
  Columns = cylinders

Custom Graph Facet Grid feature: 2D grid created with origin (rows) and cylinders (columns). Graphs arranged for each combination of origin and cylinder count for complex comparisons

Graphs are arranged for each combination of cylinder count and origin.

Scales - Scale Control

Logarithmic Scale

Logarithmic scale is effective when comparing data by ratios or when distributions are heavy-tailed.

Scales: x = log

Custom Graph Scales adjustment: scatter plot with logarithmic scale applied to X-axis. Effective when data range is wide, displaying both small and large values clearly

The logarithm is not defined for values of zero or below, so applying it to a column that contains such values suppresses the plot and shows an error identifying the affected axis, column name, and count.

Color Scale

You can specify which colors to use with color scales.

Different palettes are available for continuous and categorical variables. See Custom Graph Reference for the full list.

Sequential and Diverging scales work only with numeric columns. The scale type selector appears in the Fill aesthetic settings. When one is selected for a non-numeric column, MIDAS gives priority to the column type, renders the aesthetic with a categorical scale, and shows a warning. When a Diverging palette is also selected, the palette is incompatible with the categorical scale, so an error is shown and the graph is not rendered.

Here we change the measurement scale of cylinders to Ordinal before plotting. You can change the scale by right-clicking a column in Data Table and choosing Edit Scale of Measurement.

Aesthetics: x = weight, y = mpg, color = cylinders
Scales: Palette = Viridis

Custom Graph Scales - color palette: scatter plot with Viridis palette applied to cylinders. Sampling the palette evenly across the full range for the number of categories produces monotonic lightness changes that convey the ordinal order

Viridis is a perceptually uniform palette that accommodates color vision diversity. When applied to an ordinal variable like cylinders, it samples the palette evenly across the full range for the number of categories, so lightness changes monotonically along the order. Plasma, Inferno, and Magma offer the same property.

Category order depends on the column type. Numeric columns are sorted in ascending numeric order, ordinal enum columns follow the enum definition order, and all other columns are sorted as strings in ascending order.

Shape and Linetype Scales

For shape and linetype you choose which markers and line styles are assigned to categories. The settings appear in the layer's Aesthetics section as soon as you map a column. The first entry in the list goes to the first category.

Aesthetics: x = weight, y = mpg, shape = origin
Scales: Shapes = Square, Triangle, Diamond

Custom Graph shape scale: scatter plot assigning Square, Triangle, and Diamond to the three origin categories, with the legend listing each shape and category name

When there are more categories than assigned shapes, the shapes repeat from the start of the list. Different categories are then drawn with the same shape and cannot be told apart, so this situation is reported as a warning above the graph. The same warning appears when the list assigns one shape to more than one row, because the categories in those rows also share a shape. Linetype behaves the same way, and only Solid, Dashed, and Dotted exist, so four or more categories cannot be drawn apart by linetype.

The legend title can be set separately for shape and linetype. When it is unset, the column name is used.

Size and Alpha Scales

For size and alpha you specify the range of drawn values with a minimum and a maximum. The column minimum maps to the range minimum, and the column maximum to the range maximum. The defaults are 2px to 20px for size and 0.2 to 1 for alpha. On Text layers the range sets the font size in pixels; when it is unset, the font size runs from 8px to 24px.

Aesthetics: x = weight, y = mpg, size = horsepower
Scales: Size Range = 4, 30

Custom Graph size scale: scatter plot mapping horsepower to point size with the drawn size range widened to 4px through 30px

MIDAS draws no legend for size or alpha. To read the value behind an individual point, set the Tooltip Mode to Custom and add that column as a field.

Geometry/Statistics Reference

See Custom Graph Reference for a complete list of Geometries and Statistics.