ggplot2 facet_grid() in R: Grid Layouts With Free Scales

The facet_grid() function in ggplot2 splits one plot into a 2D matrix of panels defined by a row variable and a column variable. Rows and columns become coordinates, so cross-group comparison along either axis is read at a glance.

p + facet_grid(rows = vars(drv))                     # rows only
p + facet_grid(cols = vars(cyl))                     # columns only
p + facet_grid(drv ~ cyl)                            # rows by cols
p + facet_grid(drv ~ cyl, scales = "free_y")         # free y per row
p + facet_grid(drv ~ cyl, space = "free_x")          # column width by data
p + facet_grid(drv ~ cyl, labeller = label_both)     # "drv: 4" labels
p + facet_grid(drv ~ cyl, margins = TRUE)            # add total panels
p + facet_grid(drv ~ cyl, switch = "y")              # strips on the left

What facet_grid() does in one sentence

facet_grid() lays out panels on a strict 2D matrix where rows are the levels of one variable and columns are the levels of another. It is the cross-tabulation primitive for ggplot2: same geoms, one panel per (row_level, column_level) cell, including empty cells.

Use it when you want to compare a relationship across two categorical splits at once. For one-variable faceting, prefer facet_wrap(); for cross-tabulation, facet_grid() is the right tool.

Syntax

facet_grid() is a layer added to a ggplot() call that takes either a formula (rows ~ cols) or the explicit rows/cols arguments. Optional arguments control scale linkage, panel sizing, labels, strip placement, and margin totals.

The full signature:

facet_grid(rows = NULL, cols = NULL, scales = "fixed", space = "fixed",
           shrink = TRUE, labeller = "label_value", as.table = TRUE,
           switch = NULL, drop = TRUE, margins = FALSE, axes = "margins",
           axis.labels = "all")

You can pass rows and cols as vars(varname), or use the older formula form rowvar ~ colvar. Use a dot for "no faceting on this axis": facet_grid(. ~ cyl) is columns only; facet_grid(drv ~ .) is rows only.

Seven common patterns

1. Basic facet_grid by rows and columns

facet_grid(drv ~ cyl) creates a matrix with one row per drv level (4, f, r) and one column per cyl level (4, 5, 6, 8). Empty cells (e.g. rear-wheel-drive 4-cylinder cars) are kept and stay blank, which is the defining behavior of grid faceting.

2. Rows only or columns only

A dot on either side of the formula means "no faceting on this axis". drv ~ . gives one row per drv level, stacked vertically. Swap to . ~ drv for the same panels arranged horizontally.

3. Free scales per row or column

Unlike facet_wrap(), where scales = "free_y" frees each panel independently, facet_grid() frees each row (with "free_y") or each column (with "free_x") but keeps the linked axis consistent within the row or column. This preserves the grid's row-vs-row and column-vs-column readability.

4. Proportional panel sizes with space

The space argument is unique to facet_grid(): panels can take width or height proportional to their data range instead of every panel having the same size. Set space = "free_x" with scales = "free_x" so wider data ranges get wider panels, which avoids whitespace and visual distortion.

5. Marginal totals

margins = TRUE appends an "(all)" row, an "(all)" column, and one grand-total panel. Each marginal panel re-aggregates the data across that dimension, so the bottom row shows each cylinder count pooled over drive types, and the rightmost column shows each drive type pooled over cylinders. Useful for sanity-checking that subgroup patterns match the overall pattern.

6. Custom labellers and strip position

Default labels show only the value (f, 8). label_both shows drv: f and cyl: 8. The switch argument moves strip labels: "x" moves column strips to the bottom, "y" moves row strips to the left, and "both" does both at once. This pairs well with theme(strip.placement = "outside") when an axis label would otherwise overlap the strip.

7. Programmatic faceting with vars()

The vars() form is required when faceting variables come from function arguments. Pair it with curly-curly ({{ }}) inside a function so the caller can pass bare column names: plot_by(mpg, drv, cyl). Trying to pass drv ~ cyl programmatically forces awkward string parsing; the vars() form is the supported path.

facet_grid() vs facet_wrap()

Use facet_grid() when row and column positions both carry meaning; use facet_wrap() when one variable wraps for layout. Both share scale, theme, and labelling defaults but answer different layout questions.

Rule of thumb: if you can frame the question as "how does Y depend on X across (row, column) cells", reach for facet_grid(). If the question is "show me one panel per group, please wrap them to fit", reach for facet_wrap().

Common pitfalls

Pitfall 1: empty cells when one combination has no data. facet_grid(manufacturer ~ class) on mpg produces dozens of empty cells because most manufacturers do not build every class. Filter to a balanced subset first, or switch to facet_wrap(~ manufacturer + class) to drop empties.

Pitfall 2: free scales break cross-row comparison. scales = "free_y" lets each row pick its own y range, which is great inside the row but misleading across rows. Reserve free scales for cases where the within-row distribution matters more than the across-row magnitude.

Pitfall 3: many levels make tiny panels. A 6x8 grid yields 48 panels, each too small to read. Cap each axis at 4 to 6 levels, or lump rare ones with forcats::fct_lump().

Try it yourself

Related ggplot2 functions

After mastering facet_grid(), look at:

• facet_wrap(): wrap one variable's levels into a compact rectangle
• vars(): tidy-evaluation helper for programmatic faceting
• label_both(), label_value(), label_parsed(), label_wrap_gen(): built-in labellers
• as_labeller(): turn a named character vector into a labeller
• theme(strip.text, strip.background, strip.placement): style panel strips
• coord_cartesian(): zoom inside panels without dropping points

For paginated grids, see ggforce::facet_grid_paginate(). For interactive grids, wrap with plotly::ggplotly().

FAQ

What is the difference between facet_grid and facet_wrap in ggplot2?

facet_grid() arranges panels in a strict matrix where row and column positions match levels of two variables; empty combinations stay visible. facet_wrap() arranges panels in a 1D sequence wrapped to fit the page; empty combinations are dropped. Use grid when both axes carry meaning, wrap when only one does.

How do you use free scales in facet_grid?

Pass scales = "free_y", "free_x", or "free". Inside facet_grid(), "free_y" frees the y axis per row (not per panel), and "free_x" frees the x axis per column. This preserves cross-column or cross-row comparison along the linked axis.

How do I change labels in facet_grid?

Set the labeller argument: labeller = label_both shows var: value, labeller = label_wrap_gen(width = 15) wraps long names. For full control, build a labeller with as_labeller(c("4" = "4 cylinders", "8" = "8 cylinders")) or recode the variable before plotting.

How do I add row and column totals to facet_grid?

Set margins = TRUE on facet_grid(). ggplot appends an "(all)" row, an "(all)" column, and a grand-total cell, each showing the data pooled along that dimension. Use this to verify that subgroup patterns reflect the overall pattern.

Why are there empty panels in my facet_grid plot?

facet_grid() keeps every (row, column) cell, even when no rows match. This is by design so the grid stays rectangular. To drop empty combinations, use facet_wrap(~ row_var + col_var) instead.

For the canonical reference, see the ggplot2 facet_grid documentation.