DataslopeLine Plots
relplot with kind='line' for ordered data — trends, automatic aggregation, and confidence bands.
A scatter plot shows a cloud of independent points. A line plot is different in one crucial way: it connects points with segments, and that connection makes a promise — that the space between two points is meaningful, a continuous path from one to the next. You only get to make that promise when the x-axis is ordered.
That is the whole idea behind a line plot: it is the chart for an ordered x — usually time, or a measured sequence — where you want to see a trend, a direction of travel. Reach for it when "what happens as x increases?" is the question.
What a line plot needs and shows
- Data it needs: an ordered x (numeric or time-like) and a numeric y. The order of x is what gives the connecting line its meaning.
- What it highlights best: the trend over the ordered axis — rising, falling, cyclical, leveling off. The eye follows the slope.
- What it hides or distorts: the individual observations behind an aggregated line, and — dangerously — it implies a path even where none exists if x is not really ordered.
- When it breaks: too many overlapping lines turn into unreadable "spaghetti," and an unsorted x makes the line zig-zag back on itself.
In Seaborn you draw one with the same relplot you already know, switching
kind="scatter" to kind="line".
A single line: a clear trend over time
The flights dataset records monthly airline passengers from 1949 to 1960.
Putting the ordered year on x and passengers on y gives a textbook
upward trend.
Air travel grew steadily over the decade — you read that off the slope instantly. But look closely: there are twelve rows for each year (one per month), yet you see a single line with a shaded band around it. That is not an accident. It is Seaborn doing statistics for you.
The key behavior: aggregation and a confidence band
Here is the most important thing to understand about Seaborn line plots: when there are multiple y-values at the same x, Seaborn aggregates them. By default it plots their mean as the line, and draws a translucent confidence band around it to show how uncertain that mean is.
The fmri dataset makes this vivid — it has many measurements at each
timepoint. Color by event and you get one aggregated line per event,
each with its own band.
The solid line is the mean signal at each timepoint; the shaded ribbon is a 95% confidence interval for that mean. A narrow band means the mean is well-pinned-down (many consistent observations); a wide band means it is uncertain. You did not group, average, or compute an interval — naming the columns was enough.
A line plot is often a summary, not raw data
With repeated y-values per x, the line you see is a computed mean, not your original rows. That is a feature: it turns a noisy mess of points into a readable trend with honest uncertainty attached. Just remember the individual observations are summarized away.
Controlling the band: the errorbar argument
What the band represents is up to you, via the errorbar argument:
errorbar=("ci", 95)— a 95% confidence interval of the mean (the default). It answers "how sure am I about where the mean is?"errorbar="sd"— one standard deviation of the data. It answers "how spread out are the individual observations?" — a different question.errorbar=None— no band at all, just the mean line.
Watch the band switch meaning. With "sd", it shows the spread of the raw
signal, which is typically wider than a confidence interval.
And turn it off entirely with errorbar=None when you want a clean trend
line and the uncertainty is not the point.
A confidence interval is not the data's spread
A 95% CI says where the mean probably lies; standard deviation says how far individual points scatter. They are different quantities and usually different widths. State which one your band shows — readers cannot tell them apart by looking, and confusing them overstates or understates your certainty.
Multiple lines with hue and style
Just like scatter plots, line plots take hue to color by a group and
style to vary the dash pattern of the line for a second grouping.
Here both region and event are encoded at once.
Color separates the regions; dash style separates the events — four lines that you can still tell apart. But this is also where line plots start to strain: every extra line competes for the same space.
The pitfalls: order and spaghetti
Two mistakes hurt line plots more than any others.
1. Don't draw a line between unordered categories. A line asserts
that x flows from one value to the next. If x is a set of unordered
categories — species, island, payment_method — there is no
"between," and the connecting segments invent a progression that does not
exist. That is misleading. Use a bar or point plot for unordered
categories; save the line for genuinely ordered axes.
2. Watch for spaghetti. Pile on too many groups and the lines tangle
into an unreadable knot — the infamous "spaghetti plot." Past a handful of
series, color stops disambiguating them. The fixes: plot fewer lines,
split them into panels with col=, or highlight just the one or two that
matter.
Sort your x, or the line will zig-zag
Seaborn sorts by x before drawing by default, so a tidy column is fine. But
if you ever disable sorting (sort=False) or feed pre-shaped data in a
strange order, an unsorted x makes the line jump backward and forward and
the "trend" becomes nonsense. Ordered x in, sensible line out.
Why is a line plot the wrong choice for showing average tip across the
unordered categories day = Thur, Fri, Sat, Sun when treated as plain
categories?
A line plot cannot display categorical data at all; Seaborn will error.
The connecting segments imply a continuous progression between categories that has no real meaning.
Lines can only be drawn for numeric x-values, never for dates.
A line plot would hide the individual tip amounts.
Your turn
Using the fmri dataset, draw a line plot with sns.relplot:
timepointon the x-axis,signalon the y-axis,kind="line",- one colored line per
region(usehue).
Assign the result to a variable named g.
Check your understanding
What single property of the x-variable most justifies using a line plot instead of a scatter plot?
The x-variable is stored as a string rather than a number.
The x-variable is ordered, so the segment connecting two points represents a meaningful progression.
The x-variable has more unique values than the y-variable.
The x-variable contains no missing values.
In sns.relplot(data=fmri, x="timepoint", y="signal", kind="line"), the
fmri data has many signal values at each timepoint. What does Seaborn
draw by default?
A separate line through every individual observation.
One line connecting the maximum signal at each timepoint.
A single line of the mean signal at each timepoint, surrounded by a translucent confidence band.
Nothing — it raises an error because there are duplicate x-values.
You want your shaded band to show how spread out the individual observations are, rather than how uncertain the mean is. Which argument do you set?
errorbar=("ci", 95)
errorbar="sd"
errorbar=None
hue="sd"
You plot 30 separate companies' stock prices over time as 30 colored lines on one chart, and it becomes an unreadable tangle. What is the most useful description and fix?
It is overplotting, fixed by lowering alpha until the lines fade.
It is fine as-is; readers can trace 30 colors with effort.
It is a spaghetti plot; fix it by showing fewer lines, faceting with col=, or highlighting only the series that matter.
It means the x-axis is unordered and should be melted.
You have now met both members of the relational family: scatter plots for clouds of points and line plots for ordered trends. Next we turn to a different question entirely — not how two variables relate, but how a single variable is distributed — starting with histograms.