ggplot2 stat_summary() in R: Plot Means and Error Bars

The stat_summary() function in ggplot2 computes a summary statistic per group on the fly, then draws it as a point, line, errorbar, or other geom. It is the fastest way to overlay group means, medians, standard errors, or confidence intervals on raw data, without a separate dplyr aggregation step.

stat_summary(fun = mean, geom = "point")                        # mean per group
stat_summary(fun.data = mean_se, geom = "errorbar")             # mean +/- SE
stat_summary(fun.data = mean_cl_normal, geom = "pointrange")    # mean + 95% CI
stat_summary(fun.data = mean_cl_boot, geom = "pointrange")      # bootstrap CI
stat_summary(fun.data = median_hilow, geom = "pointrange")      # median + IQR
stat_summary(fun = mean, geom = "line", aes(group = 1))         # connect means
stat_summary(fun = median, fun.min = min, fun.max = max)        # custom range

What stat_summary() does in one sentence

stat_summary() summarises y values at each unique x, then draws the result with a chosen geom. It accepts raw long-format data, applies a function like mean or median per x-group, and renders one summary per group using geom = "point", "line", "errorbar", "pointrange", or any compatible geom.

This eliminates the common two-step pattern of dplyr::summarise() followed by geom_point(). With stat_summary(), the aggregation happens inside the plot layer, the raw data stays in the data argument, and you can overlay summaries on top of raw points without merging two data frames.

Syntax

stat_summary() is a ggplot2 layer that takes either fun or fun.data, plus a geom. Use fun when the summary is a single y value (mean, median). Use fun.data when the summary is three values (y, ymin, ymax) for ranges and error bars.

The full signature:

stat_summary(mapping = NULL, data = NULL, geom = "pointrange",
             position = "identity", ...,
             fun.data = NULL, fun = NULL,
             fun.max = NULL, fun.min = NULL, fun.args = list(),
             na.rm = FALSE, orientation = NA,
             show.legend = NA, inherit.aes = TRUE)

Key arguments:

• fun: a function returning one number (e.g. mean, median).
• fun.data: a function returning a data frame with y, ymin, ymax (e.g. mean_se, mean_cl_normal, mean_cl_boot, median_hilow).
• fun.min, fun.max: lower and upper bounds when used with fun.
• fun.args: list of extra arguments passed to fun or fun.data (e.g. list(mult = 2) for 2x standard error).
• geom: the geom that draws the result. Default is "pointrange".

Six common patterns

1. Mean point per group

fun = mean collapses each cylinder group to one number. geom = "point" draws it as a single dot per x value. Wrap cyl in factor() so the x axis is discrete.

2. Mean with standard error bars

mean_se (from Hmisc, re-exported by ggplot2) returns the mean plus and minus one standard error. Two stat_summary layers combine to show the point and the bar; the order draws the bar first and the point on top.

3. Mean with 95% confidence interval

mean_cl_normal computes the normal-theory 95% CI on the mean. geom = "pointrange" renders the mean as a point with a vertical line for the CI in one layer. Pass fun.args = list(conf.int = 0.99) for a 99% interval.

4. Median with IQR

median_hilow returns the median and quantiles. With conf.int = 0.5, ymin and ymax are the 25th and 75th percentiles, giving the interquartile range. Use this for skewed data where median + IQR is more honest than mean + SD.

5. Connecting means with a line

Two layers, same fun = mean, different geom. The line traces the group means across dose; the point marks each. Color is mapped to supp, so each supplement gets its own line.

6. Bootstrap confidence interval

mean_cl_boot resamples B times (default 1000) and reports the empirical 95% CI. Robust to non-normal data. Slower than mean_cl_normal but distribution-free.

stat_summary() vs alternatives

Use stat_summary() when summaries are simple and the raw data lives in one frame. Pre-aggregate with dplyr only when summaries are complex or reused across many plots.

When to use which:

• Use stat_summary() for exploratory plots and standard summaries.
• Use dplyr::summarise() plus geom_point() when you need the aggregated table elsewhere (printed, exported, joined).

Common pitfalls

Pitfall 1: forgetting to set geom. The default geom = "pointrange" requires fun.data (three values). Pass only fun = mean with the default geom and you get a cryptic warning about missing ymin/ymax. Fix: set geom = "point" when using fun, or set fun.data = mean_se when keeping pointrange.

Pitfall 2: continuous x with too many unique values. stat_summary computes one summary per unique x. With continuous x (hundreds of unique values) you get one mean per single observation, which is just the original points. Fix: bin x first (cut() or factor()) or use stat_summary_bin() for automatic binning.

Pitfall 3: legends from constant aesthetics. Setting color = "blue" inside aes() creates a fake legend entry. Always place constant colors and sizes OUTSIDE aes(): stat_summary(fun = mean, geom = "point", color = "blue").

Try it yourself

Related ggplot2 functions

After mastering stat_summary(), look at:

• stat_summary_bin(): same idea but with automatic binning for continuous x.
• stat_summary_2d(), stat_summary_hex(): 2D analogues for heatmaps of summarised z.
• geom_pointrange(), geom_errorbar(), geom_linerange(): the bare geoms used by stat_summary's geom argument.
• stat_smooth(), geom_smooth(): fit and draw a smoothing function, a different kind of per-x summary.
• mean_se, mean_cl_normal, mean_cl_boot, median_hilow: the helper summary functions, each returning a data frame with y, ymin, ymax.

For pre-aggregated data, dplyr::summarise() followed by geom_point() and geom_errorbar() is the explicit alternative.

FAQ

What is the difference between fun and fun.data in stat_summary?

fun takes one column of y values and returns a single number (e.g. mean, median). It pairs with single-value geoms like point or line. fun.data takes the y values and returns a data frame with y, ymin, ymax. It pairs with range geoms like pointrange, errorbar, linerange. Choose fun when you want one summary value, fun.data when you want a centre plus a range.

How do I add error bars to a bar chart in ggplot2?

Layer stat_summary(fun = mean, geom = "bar") for the bars and stat_summary(fun.data = mean_se, geom = "errorbar", width = 0.2) for the bars on top. The bars and errorbars share the same per-group means because both layers compute from the same raw data. For pre-aggregated data, replace stat_summary with geom_col() and geom_errorbar().

Can stat_summary work with a continuous x axis?

It can, but you usually want stat_summary_bin(). Plain stat_summary computes one summary per unique x value. With continuous x, almost every value is unique, so the result is just the raw points. stat_summary_bin first bins x into intervals, then summarises within each bin, which is what you typically want.

How do I pass extra arguments to the summary function?

Use the fun.args argument, a list. Example: stat_summary(fun.data = mean_cl_normal, fun.args = list(conf.int = 0.99)) switches from a 95% to a 99% interval. For mean_cl_boot, use fun.args = list(B = 5000) to increase bootstrap iterations. Anything fun or fun.data accepts can go inside this list.

Why does my stat_summary plot show a warning about missing ymin?

You used the default geom = "pointrange" with only fun. The default geom needs y, ymin, ymax. Either switch to geom = "point" (needs only y) or supply fun.data = mean_se (which produces all three). The error is silently a warning, but the plot will skip the range and look incomplete.