A graph with no error bars is an opinion

Reading uncertainty, and what its absence usually means.

A graph with no error bars is an opinion because a single point plotted on its own claims a precision the underlying measurement almost never has, presenting one possible reading as though it were the only one, when every real measurement carries some honest range of uncertainty around it. Reading a chart well means treating the missing bars as a question rather than as a reassurance.

A forty-minute drive that takes thirty to fifty-five

A journey quoted as taking forty minutes sounds like a single, reliable figure, but anyone who has made the same drive repeatedly knows it can run from thirty minutes on a clear evening to nearly an hour when traffic is heavy. The honest description of that journey is a range. Leaving for an appointment on the bare forty-minute figure, with no sense of how far the real time might drift, risks arriving needlessly early or late depending on which end of the range the day happens to land on.

The quoted time is incomplete in exactly the way a graph with no error bars is incomplete, presenting the middle of a range as the whole story. A journey planner willing to say forty minutes, usually between thirty and fifty-five, gives the traveller something to plan around, while one that insists on a single figure quietly hands the risk of the hidden range to whoever trusted it.

What the bar is drawn from

An error bar is a small mark added to a plotted point showing the range within which the true value probably sits, drawn from the same kind of repeat measurements that reveal how much ordinary scatter a test produces. The scatter is present in the underlying measurement whether it is shown or not, and leaving off the bar simply removes the reader's only way to see it. A bar can show the spread of the individual readings or the narrower uncertainty in their average, and a good chart says which, since with nine repeats the second is a third the width of the first. Either way, the bar turns a dot into a claim about how far the reading could honestly have wandered.

This matters enormously once two points are being compared. Two points that look clearly different as bare dots can turn out to have overlapping error bars once the uncertainty around each is drawn in, meaning the apparent difference could easily be ordinary scatter. That is the most common way a bare chart misleads: a reader has every reason to assume the visible gap between two points reflects a real difference, and nothing on the page suggests otherwise. Two points a finger's width apart can tell entirely different stories depending on whether their ranges are each a hair wide or each wider than the gap.

A well-drawn error bar is therefore a sign of care, showing that whoever made the chart checked how much the measurement varies before presenting the first number that came back.

Asking for the spread before a costly decision

Asking to see the error bars, or the underlying spread of repeats, before accepting that two results differ costs nothing and catches a great deal of false confidence. A report claiming that a redesigned part performs better than the original, backed by one average result from each, is making a claim that a few repeats on both sides could confirm or quietly dissolve. Whether to trust it should hinge on seeing that comparison, whatever story is attached.

This becomes especially important where a decision carries real cost: choosing between two suppliers, approving a design change, retiring a process that has worked for years. Those are the moments when missing error bars can turn an uncertain result into a confidently stated fact that nobody checked.

Exact counts need no bars

Some charts are honest without error bars. A plot of an exact count, such as how many units shipped on a given day or how many parts passed inspection, involves no estimate drawn from a sample, so there is no repeat-to-repeat scatter for a bar to represent. Adding one would invent doubt where none exists and misrepresent the data.

The caution belongs to charts built from measurements or samples, anything that would come back at least slightly different if repeated. Those are the cases where a bare point substitutes one convenient reading for the range the underlying process actually produced, and where a reader has every right to ask how wide the uncertainty around each point is before trusting the gap between them. Telling an exact count from an estimated measurement is a small skill worth having before either demanding bars everywhere or dismissing them as decoration.