Never connect a categorical axis with lines

What it is

“Categorical” in the title means a nominal variable with no intrinsic order or distance. Connecting regions, brands, or product types turns an arbitrary arrangement into apparent direction, turns, and slope; reorder the categories and the trend changes. Unconnected dots or bars usually support the intended magnitude comparison. The title is not a ban on every line between categories: ordered stages, paired conditions, and genuine relational edges can be connected, but the line must encode defined sequence, pairing, or network structure rather than invented continuous change.

Why it happens

A segment gives adjacent marks both connectedness and direction. Readers may interpret vertical movement as process, horizontal spacing as distance, and slope as rate. Nominal categories supply none of those structures, so the path shape follows the author's ordering. Even when a line is intended only to help match one series, it adds semantics; labels, alignment, common region, or interactive highlighting may support matching with less risk of implying a process.

Studying it

Generate several orders of the same nominal data and compare connected, dot, and bar versions. Ask for category comparison, ranking, series matching, and verbal takeaways; measure error, time, and use of process terms such as rising, falling, or fluctuating. When a connection has an explicit relational meaning, separately test whether readers distinguish category comparison, paired change, and network edge. Reordering sensitivity is diagnostic, not a sufficient verdict for ordered or paired designs whose sequence is part of the study meaning.

Where it stops holding

Ordinal categories have order but may lack equal intervals. A connection can emphasize sequence without supporting rate. Repeated-measure conditions arranged by experimental progression may connect each participant's paired observations; the pairing should be stated, and the result is not a continuous process between categories. Parallel-coordinate paths join one record across dimensions, and network edges encode relations; neither is a line-chart trend. Date labels positioned by real time form a temporal domain even though their labels look categorical.

Applying it

• Start nominal horizontal variables with dots or bars. Order categories for the analysis question and verify that the takeaway does not depend on an unexplained arrangement.
• If a connection is necessary, state whether it represents sequence, within-object pairing, or a relation edge, and avoid slope or interpolation language for nominal intervals.
• Give lines, endpoints, and relations programmatic names and structure. A nonvisual representation should identify source, target, direction, and pairing rather than merely announce coordinates.
• Compare reasonable category orders in task testing. Remove connections, or use facets, labels, and highlighting, when participants infer a process trend from arbitrary order.

Related

• Same group: U2.03.1 Line charts suit trends over ordered variables · U2.03.3 Beyond a few lines, tracking collapses
• Adjacent: U1.13.2 Connectivity makes a line read as one continuous process · U2.02.1 Bar charts suit magnitude comparison across categories
• Search terms: nominal axis · categorical line chart · spurious trend · paired observations