Level 3 · Six lines or six small charts

The spaghetti problem

Chart A puts six categories on one chart. Each line has its own colour, and the legend maps colours to names. To follow one category, your eye has to find its colour, trace it through every crossing, and keep the legend in mind. With two or three lines that works. With six it turns into spaghetti: you see the overall shape (everything jumps in November) and lose each line’s story.

Colour also runs out. Six colours that people with colour-blindness can tell apart, on a dark background, is about the limit. A seventh series would need to fold into “Other”.

Small multiples

Chart B draws the same data as six small charts, one per category, in the same order every time. Three rules make it work:

• Same axis on every panel. A $60k bump looks the same size in every panel, so you can still compare categories. A panel with its own axis would make a small category look as big as a large one.
• Same x range and layout. January is always on the left edge and December on the right.
• The panel title names the series, so there is no legend to decode and no colour to match.

The reader now compares shapes, which is what eyes are good at. Five panels rise at the right edge. Outdoor has its hill in the middle.

Building the data

A small-multiples chart needs data in long form: one row per month per category.

monthly = (
    lines.assign(month=lines["order_date"].dt.to_period("M"))
    .groupby(["category", "month"])["revenue"].sum()
    .reset_index()
)

Most chart tools then take a “facet” or “row/column” setting on category. In pandas with matplotlib, monthly.pivot(index="month", columns="category") gives one column per panel.

When one chart is still better

Small multiples hide exact crossings. If the question is “when did Lighting overtake Kitchen?”, two lines on one chart answer it directly. Use one chart for two or three series whose relation is the point, and small multiples for many series whose individual shapes are the point.

Common mistakes

• Letting each panel pick its own axis. It makes every category look equally large and hides the real differences in size.
• Ordering panels at random. Sort them by total or keep a fixed business order, so readers learn where to look.
• Adding more colours to rescue spaghetti. More colours make the legend longer, not the chart clearer.