Col vs Row

[Jasmine Maddox]

I always get confused when writing up my results how to accurately describe my outputs when I use column versus row (see attached photos). Could some provide clarity? Which is "better" or the most "accurate" to use?

[Leonardo Guizzetti]

Your image is mislabeled. Row percentages add to 100% across rows, and likewise for column percentages. These is no such thing as one being better or more accurate. Which you use depends on what you are trying to say. In your example, a row-wise percentage tells you the proportion with diabetes within mother age category. The column-wise percentages tell you about the distribution of mother age categories within diabetes status.

[Nick Cox]

Often, but not always, what you most want to compare are outcomes given predictors, not predictors given outcomes.

It can be easier to compare numbers within (namely, up and down) columns than across rows (namely, left to right).

In primary school (ages 5 to 11) I was taught to put numbers in columns and add them downwards and check going upwards. I dare say that was a common experience, but perhaps small children are now expected just to use their phones. (Mental arithmetic was where nothing but the answer could be written down and mechanical arithmetic was where you could show "workings" on your paper.)

Sometimes those points come together without contradiction. Other times there isn't even a clear distinction between the roles of variables, or the goal is descriptive, rather than analytical, to show association, or to lay out marginal and conditional distributions, or other summary measures.

Otherwise, table layout can be determined as much as by convenience and convention as by any kind of compelling scientific, statistical or psychological logic.

Confession: I don't often think hard about which way round a table should be. But I have often tried a table one way round and then reversed rows and columns whenever what I tried first was clearly awkward. I am especially reluctant to adopt split tables with two or more bunches of columns.

The British statistician A.S.C. Ehrenberg [Andrew] wrote much about tables. Following his advice would lead to the rejection of many dissertations, theses, and journal papers, as in his terms most tables are far too complicated. The word ascerbic might have been invented for him, judging by some of his writing, but when I once met him he was perfectly civil too. Look for books or papers by him with "data reduction" or "numeracy" in the title.

Particular disciplines or journals may have guidelines or even rules that should imply a decision.

It's interesting that there seems to be a much stronger convention in graphics that whatever you think is the outcome or response -- which you are trying to describe, explain or predict -- should go on the y (vertical) axis. This principle can even be reconciled with quite different displays of univariate distributions, as variously the outcome of interest is the frequency or density (histograms), the cumulative probability or survival probability, or the set of quantiles or order statistics (quantile plots).

But, but, but:

1. Earth and environmental scientists and archaeologists often show height above or depth below surface on the vertical axis, and what could be more vertical?

2. In some of the same fields, time (whether just ordered or measured) is often shown vertically. The logic seems to be that older things (sediments, fossils, pots, broaches, skeletons) are typically (buried) below younger things.

3. Many bar charts should be laid out horizontally so that category labels can be easily readable. It is sometimes puzzling, sometimes diverting to see struggles with label text size or angle in attempts to show readable vertical bar charts (column charts, in a terminology that lingers).