A slide and a page are different machines. On a slide, the title sits above the exhibit and the reader meets them together. In a document, prose that has already made the argument points to the exhibit, the reader arrives from a sentence, and the caption is what they read at the exhibit.
What makes a good chart caption?
That changes the caption's job. It carries the claim, as an action title does. It also carries the exhibit number the prose referred to, and the source line that makes the numbers checkable. A working caption for a scatter of fourteen US industrials reads: "Exhibit 4. Companies earning higher returns on capital trade at higher multiples of it; Meridian Fluid, at 22.0% return and 1.10 times invested capital, trades 57.2% below the fitted relationship. Source: synthetic dataset, latest reported fiscal year."
Three things are doing work there. The claim uses the association verb and avoids a causal one, because a cross-sectional scatter with an R-squared of 0.69 doesn't establish direction. The exception is named with its two numbers, because the exception is why the exhibit is in the document at all. And the source line is there, which in a research context is the difference between an exhibit and an assertion.
Advertisement
The test for a document's set of exhibits is the caption-only read. Someone who reads nothing but your captions, in order, should get the argument. It's the written-document version of the read-only-the-titles test a deck gets, and it fails the same way. A run of captions saying "Exhibit 3. Revenue by segment" and "Exhibit 4. Margin by segment" tells a reader that segments exist and nothing else.
One number in that caption is worth following, because it shows something an exhibit can carry and prose can't. The fitted line through the fourteen companies has a slope of 0.1179 times invested capital per point of return and an R-squared of 0.6887. Remove Meridian Fluid and refit, and the R-squared rises to 0.9932. So thirteen of the fourteen companies sit almost exactly on a line and one doesn't. That's a fact about the shape of the data, and a correlation coefficient in a sentence would have buried it. The scatter shows it in one look, and the caption's job is to say what the look is for.
How should you number and place exhibits?
Numbering and placement follow the medium. Number exhibits in the order the prose refers to them, and place each one after its first reference. A document that refers to Exhibit 7 before Exhibit 3 has a structural problem, and the numbering has just exposed it.