ggplot2 geom_smooth() in R: Trend Lines With Examples

The geom_smooth() function in ggplot2 fits a smooth trend line through (x, y) data with an optional confidence band. Use method = "lm" for linear regression, "loess" for local smoothing, or "gam" for generalized additive splines.

ggplot(df, aes(x, y)) + geom_point() + geom_smooth()                 # default loess
ggplot(df, aes(x, y)) + geom_point() + geom_smooth(method = "lm")    # linear regression
ggplot(df, aes(x, y)) + geom_smooth(method = "lm", se = FALSE)       # no CI band
ggplot(df, aes(x, y, color = group)) + geom_smooth(method = "lm")    # one line per group
ggplot(df, aes(x, y)) + geom_smooth(method = "lm", formula = y ~ poly(x, 2))
ggplot(df, aes(x, y)) + geom_smooth(method = "gam")                  # GAM splines
ggplot(df, aes(x, y)) + geom_smooth(method = "loess", span = 0.3)    # tighter loess

What geom_smooth() does in one sentence

geom_smooth() fits a trend line to (x, y) data and optionally draws a 95% confidence band around it. The method argument picks the model: linear regression (lm), local polynomial (loess), generalized additive (gam), or any custom modeling function.

For exploratory analysis, geom_smooth() is the fastest way to add a quick statistical summary to a scatter plot. For publication, fit the model separately with lm() or gam(), extract predictions, and plot via geom_line() for full control over confidence levels, prediction intervals, and styling.

Behind the scenes, geom_smooth() calls a stat that subsamples your data, fits the chosen model, predicts y values across a regular grid of x, and connects those predictions with a line. The confidence band is computed from the model's standard errors. None of this is visible in the API; you just get a clean line on top of your scatter plot.

The smoothness of the resulting curve depends on the method: lm gives a perfectly straight line; loess and gam adapt to the data shape with parameters you can tune. Default settings work well for exploratory plots; for publication, always specify the method and any tuning parameters explicitly.

Syntax

geom_smooth() requires aes(x, y). Other arguments tune the model.

The full signature:

geom_smooth(mapping = NULL, data = NULL, stat = "smooth", position = "identity",
            ..., method = NULL, formula = NULL, se = TRUE, na.rm = FALSE,
            orientation = NA, show.legend = NA, inherit.aes = TRUE)

Six common patterns

1. Default smooth (loess for small data, gam for large)

The simplest case: loess curve fitted through the points, with a 95% confidence band. ggplot prints a console message indicating which method it chose.

2. Linear regression line

method = "lm" fits ordinary least squares and draws the line. The shaded band is the 95% confidence interval for the mean prediction.

3. Linear fit without confidence band

se = FALSE hides the confidence band. Use this when the band is misleading (small samples) or when you want a cleaner visual.

4. One smooth per group

Mapping color = drv automatically groups the smooth. Each drv gets its own regression line. Helpful for comparing trends across categories.

5. Polynomial fit

formula = y ~ poly(x, 2) fits a degree-2 polynomial. For higher-order, change the integer (poly(x, 3), etc.). Use splines::ns(x, df = 4) for natural splines.

6. Tighter LOESS span

span controls how much data is used at each x point. Default 0.75 is smooth; 0.3 is more responsive to local features. Lower span = wigglier fit.

This separation matters in practice. A team might prototype with geom_smooth(method = "lm") and discover a useful linear trend. The actual analysis then runs lm() separately to get coefficients, residuals, and diagnostics. The final report uses both: a fitted-line plot for the visual story and a coefficient table for the statistical claim. geom_smooth() is the entry point, not the destination.

A note on confidence intervals in geom_smooth

The shaded band drawn by default with se = TRUE is the 95% confidence interval for the MEAN PREDICTION at each x. It tells you "the average y at this x is in this band with 95% confidence", assuming the model is correct. It is NOT a prediction interval for individual observations, which would be wider. For prediction intervals, fit the model with lm() and use predict(model, interval = "prediction") to extract the wider band, then plot via geom_ribbon().

The CI band's width depends on the model's standard errors at each x, the chosen significance level (default 0.95), and the spread of x values around that point. It is generally narrowest near the center of the x range and widens toward the extremes, which matches how a regression model is most certain near the mean of x.

geom_smooth() methods comparison

Choose method based on what you want to show.

When to use which:

• Use lm when you want to show or test a linear trend.
• Use loess for exploratory nonlinear smoothing on small data.
• Use gam for large data or when you need spline-based nonlinearity.

Common pitfalls

Pitfall 1: forgetting that geom_smooth's confidence band is for the MEAN PREDICTION, not for individual points. A new observation at a given x has wider uncertainty than the band shows. The band tells you "the mean of y at this x is in this range with 95% confidence", not "any new point will fall in this range".

Pitfall 2: relying on the default method. Default switches between loess and gam at N=1000. Adding 50 rows of data can change the visual. Always specify method for reproducible plots.

Pitfall 3: loess fails on very large data. Loess is computationally expensive. With 100K+ rows, it can be slow or run out of memory. Switch to gam for large data or sample the data first.

Try it yourself

Related ggplot2 functions

After mastering geom_smooth(), look at:

• stat_smooth(): alternative spelling; same function
• stat_summary(): custom summary statistics on top of points
• geom_line() plus model predictions: full control over the fitted line
• lm(), glm(), gam(): fit the model separately for full diagnostic control
• ggeffects package: extract and plot model predictions cleanly
• geom_ribbon(): manual confidence bands when you want full control

For mixed-effects models, fit with lme4::lmer() and use ggeffects::ggpredict() to extract predictions for plotting.

FAQ

How do I add a regression line in ggplot2?

Use geom_smooth(method = "lm") after geom_point(). For just the line without the confidence band: geom_smooth(method = "lm", se = FALSE). For higher-order: formula = y ~ poly(x, 2) for quadratic.

What is the difference between lm and loess in geom_smooth?

method = "lm" fits a straight line via ordinary least squares. method = "loess" fits a smooth local curve that adapts to the data shape. Use lm for linear; loess for exploratory nonlinear.

How do I remove the confidence band in geom_smooth?

Add se = FALSE: geom_smooth(method = "lm", se = FALSE). The band is hidden but the line stays.

Why is my geom_smooth slow?

Loess (the default for small data) is O(N^2) and slows on large datasets. Solutions: switch to method = "gam" for large data, or sample the data first via slice_sample(prop = 0.1).

How do I fit a separate trend line per group in ggplot2?

Map a categorical variable to color, linetype, or group: aes(x, y, color = grp) + geom_smooth(method = "lm"). ggplot will fit one regression per unique grp value.