Linked exploratory data analysis

How LinkEDA preserves case identity and selection across native views.

One data set · one set of row identities

Linked exploratory data analysis

LinkEDA treats selection as analytical state. A point is not merely coloured inside one chart: it represents a row in the registered data set, and that row identity can be recognised by every compatible view in the same linked group.

Selection across views

When a plot is created from an active data set, it receives the data set’s linking group and a mapping from each graphical mark back to original row indices. The portable core stores the selected indices. A click or brush updates that shared state, not a private selection inside the window.

The update then follows this path:

1

Hit testing identifies cases

The plot converts the pointer position to plot coordinates and determines which point, line, bin or bar segment was hit. Aggregate marks resolve to all rows that contributed to them.

2

The selection operation is applied

Replace, add, subtract and toggle operations update the shared set of original row indices. Brushing applies the same logic to every case inside the rectangle.

3

Linked views are notified

Scatterplots, boxplots, histograms, bar charts, trellis panels and the data sheet that share the group receive the new state and redraw their corresponding cases.

4

The theme supplies the appearance

Selection is shown using the theme’s selected and unselected colours. Linking does not force a hard-coded colour, so the same selection remains consistent with the active visual theme.

This means, for example, that selecting a point in a scatterplot can highlight the same car in a second scatterplot, its point in a grouped boxplot and its row in the data sheet. Clicking a bar segment is different only in scale: because the segment summarises several rows, all contributing observations become selected.

What linking does—and does not—share

Linking shares case selection. It does not force plots to use the same variables, coordinate system, scale or aggregation. Two windows can therefore show different statistical representations while referring to the same observational units.

Plots created with linking disabled, or from separately registered data sets without a common mapping, do not participate in the selection. Missing values can also mean that a selected row has no visible mark in a particular plot even though the row remains selected in the data set.

Selection operations

  • Replace makes the newly hit cases the selection.
  • Add keeps the existing selection and includes new cases.
  • Subtract removes hit cases.
  • Toggle reverses the state of hit cases.
  • Clear, Invert and All operate on the current linked universe.

The same operations are available to point selection and brushing where the plot supports them.

From selection to analysis

A transient selection can be given a name and retained as an analysis scope. The named scope then appears in the scope menu and can be reused after the on-screen selection changes. This distinction matters for reproducibility: result windows can state which named set of rows was analysed instead of ambiguously referring to “the selected cases”. Saved scopes can also be combined or used to create a subset through the Data menu.

Note

Linked highlighting is exploratory evidence, not an inferential test. If an apparent subgroup matters, define an appropriate analysis scope and fit a model or test that addresses the question explicitly.