In short
An APA 7 correlation table reads in one pass because it always has the same shape: bold table number and italicized title above; variables numbered (1, 2, 3…) down the first column; M and SD in the next two columns; and only the lower triangle of the matrix, because the upper half repeats exactly the same coefficients. The diagonal takes a dash or an empty cell — 1.00 is a tautology that eats space. Many organisational and health psychology journals want the reliabilities in there instead, in brackets — (.89), (.84) — which APA neither requires nor forbids, but if you do it the note has to say so in plain words. Coefficients are given to two decimals with no leading zero (r = .52, r = −.18), M and SD do keep theirs, and the asterisks are declared on the last line of the note: *p < .05, **p < .01, ***p < .001. If Pearson and Spearman coefficients share the table, say which is which with a superscript letter or by giving each one its own triangle.
The correlation table looks like the easy one — four variables, six coefficients — and it is the one that comes back marked up most often. What the editor sends back is nearly always the same list: the full square matrix pasted from the SPSS viewer with both halves and ones down the diagonal, alphas slipped in among the coefficients without a word to say they are alphas, means with no standard deviations, and a note that does not even give the sample size. This is the table treated as an object with rules of its own: what occupies each position, what belongs on the diagonal, where M and SD go, what the asterisks actually declare, and what to do when half your variables call for Spearman. For the wider picture of the whole section, start with how to report results in APA 7.
What occupies each position
The structure is not something you decide per paper. You learn it once and repeat it: bold table number on its own line, flush left; italicized title underneath; first column listing the variables numbered and named ("1. Emotional exhaustion"), abbreviated there if the full name will not fit, with the note spelling it out. The matrix columns are headed with the number only, never with the name again — the name is already on the left, and repeating it turns the table into a wall of text. Between the names and the matrix sit M and SD.
Take a study of 214 nursing students. Table 1 would carry four rows — 1. Emotional exhaustion (M = 3.12, SD = 1.04), 2. Perceived stress (M = 21.40, SD = 6.31), 3. Sleep quality (M = 7.86, SD = 3.22) and 4. Academic engagement (M = 4.05, SD = 0.97) — and six coefficients: r12 = .52, r13 = .41, r14 = −.18, r23 = .34, r24 = −.14 and r34 = −.09. Four variables give six pairs; six variables give fifteen, and at that point the table stops decorating the text and becomes the only way to tell the story.
The formatting has four rules and none of them bends. Three horizontal rules (above the headers, below them, and at the close) and not one vertical rule. Two decimals throughout. A leading zero on M and SD, which can exceed 1, and no leading zero on r, which cannot. And the variables in the order your hypotheses use, not the order the questionnaire measured them in or the order the software printed them.
Past eight or nine variables the table stops fitting in portrait. Then you turn it landscape, split it in two, or move it to supplementary material — you do not shrink the type to seven point or cut the decimals to one to squeeze it in. Building it by hand in Word is half an hour of fighting cell borders: the APA correlation table generator takes the matrix exactly as SPSS, R, jamovi or Excel produce it — full or triangular, comma decimals, asterisks stuck to the numbers — and returns the table with M and SD up front, the lower triangle, the dash on the diagonal and the note at the foot, ready to paste into Word with the rules already set.
Half a matrix: the lower triangle, and why the upper one is waste
A correlation matrix is symmetric: the correlation of 1 with 2 is exactly the same number as the correlation of 2 with 1, not a similar one. Printing both halves means printing every value twice and inviting the reader to check whether they agree. The convention is to keep the lower triangle, the half below the diagonal.
Two oddities follow from that, and neither is a formatting slip. The first row carries no coefficients at all: variable 1 shows M, SD and nothing else, because there is no earlier variable to correlate it with. And the last column never exists: with four variables, the matrix headers run 1 to 3. If your table has a column headed "4" standing empty, delete it.
Both oddities change the moment the diagonal carries something. With the alphas there, row 1 does have one filled cell — variable 1's alpha — and the headers run 1 to 4, because the last variable's alpha needs a column of its own. With the diagonal empty or dashed, that column really is waste. It is a one-glance check: count how far the headers run and look at what you decided to put on the diagonal.
The upper triangle is used only when it holds different information, and then the note has to say so. The two versions you see in print are two groups (women's coefficients below the diagonal, men's above) and two time points. It is elegant and saves a whole table, but it doubles the reading load: without the note in front of them, no reader can tell which half they are looking at. What never gets published is the SPSS viewer block with both halves, ones down the diagonal, a "Sig. (2-tailed)" row and an "N" row under every coefficient. That is software output, it takes three times the space and it says the same thing. The N goes once, in the note; the p values go into the asterisks or into a column of their own.
The diagonal: a dash, ones, or the alphas
By default nothing goes on the diagonal: an em dash, or an empty cell. Writing 1.00 four times fills space with a definitional truth — a variable correlates 1 with itself whatever your data do — and it tempts the reader to read the diagonal as a result.
The habit of putting reliabilities there comes from organisational and work psychology journals and has spread across much of health psychology: each scale's alpha in brackets in its own cell — (.89), (.84), (.78), (.91) — plus one sentence in the note. The benefit is genuine. The reader can see at a glance whether a weak correlation might be a measurement problem rather than a theoretical one, without going back three pages to hunt for the alphas.
The case against is genuine too. An alpha is not a correlation, and dropping it into a grid of correlations makes the reader switch metric inside the same lattice. With variables that are not scales — age, number of night shifts, a single item — the diagonal ends up half empty and the gaps need explaining. The clean alternative is an alpha column next to M and SD: it reads just as fast and it does not put two different quantities in the same cell.
APA 7 neither mandates nor bans any of the three options; what it does demand is that the table stand alone. If reliabilities sit on the diagonal, the note says so literally ("Cronbach's alpha coefficients appear in parentheses along the diagonal"), and if the manuscript reports omega rather than alpha, it says omega. How each one is computed and reported is in how to report Cronbach's alpha in APA 7. The rule people forget: pick one option for the whole manuscript. Two correlation tables with different diagonals is a red flag to any reviewer.
Means and standard deviations: left-hand columns or bottom rows
Both layouts are acceptable and both get published. The usual one, and the one the sample tables in Chapter 7 of the Publication Manual use, puts M and SD as the first two columns, between the variable name and the matrix, so the descriptive and the coefficient sit on the same line of reading. The other puts them as the last two rows, under the matrix, aligned with the numbered columns.
The bottom-rows version is not a whim. With ten or twelve variables, columns are precious and taking two away from the matrix wrecks the page width. The cost is that the reader has to drop to the foot of the table every time they want to know what scale the variable they are reading is on. With four or five variables it is not worth it; with twelve, sometimes it is.
Wherever they sit, a mean with no range tells the reader nothing. M = 3.12 says nothing about whether exhaustion is high until you know the scale runs 0 to 6, and M = 21.40 says nothing until you know the PSS-10 runs 0 to 40. Put the possible range in the general note, or in a "Range" column if you have room. It is one line and it heads off the most predictable question a reviewer can ask.
When n is not the same in every cell — pairwise deletion, a questionnaire that dropped out halfway through data collection — say so. A superscript letter on the affected variable with its specific note will do, or one sentence in the general note: "ns range from 198 to 214 because of pairwise deletion." It costs a dozen words and it is one of the things reviewers ask for most. The in-text sentence that accompanies the table, with the coefficient, its degrees of freedom and its p, is built the way how to report a correlation in APA 7 sets out.
The asterisks, and what the note has to say
The convention is the familiar one: *p < .05, **p < .01 and ***p < .001, with no space between the asterisk and the p, the p italicized and the values without leading zeros. They go on the last line of the note, after the general and specific notes, and you declare only the levels the table actually uses: if no coefficient reaches .001, the third asterisk is not declared.
It is worth remembering what an asterisk measures, because a matrix shows it more clearly than anything else: it measures sample size as much as effect size. With N = 214 the significance threshold sits at r = .134, so r = .14 is the first two-decimal value that earns a star: in the example table r24 = −.14 gets one (p = .041) while r34 = −.09 gets nothing (p = .190), five hundredths apart. In a sample of a thousand people, r = .07 — which does not explain half a percentage point of variance — would come out starred (p = .027).
That is why a growing number of journals prefer the exact p value in brackets under each coefficient and no asterisks at all: the number informs, the symbol does not. Confidence intervals are what APA 7 genuinely pushes, but a matrix carrying an interval for every pair becomes unreadable past five variables; the sensible compromise is to give the interval for the two or three correlations your hypotheses rest on, in the text or in a separate column, and leave the rest as coefficients.
The note under a correlation table has three layers in a fixed order — general, specific, probability — and the general note has room for five things it needs: the N, what each abbreviation stands for, which coefficient was used, what is on the diagonal, and how any dichotomous variables were coded, so the sign of a point-biserial correlation can be read. For the example above it would run: "Note. N = 214. EE = emotional exhaustion; PS = perceived stress; SQ = sleep quality; AE = academic engagement. All coefficients are Pearson's r. Cronbach's alpha coefficients appear in parentheses along the diagonal. Higher SQ scores indicate poorer sleep quality."
Pearson and Spearman in the same table, without misleading anyone
In clinical data this comes up constantly: two reasonably symmetric variables, one skewed with a floor at zero, and a five-point scale nobody dares analyze as continuous. Computing Pearson for some pairs and Spearman for others is perfectly defensible. Putting them in the same grid without saying so is not: the reader assumes the whole matrix uses one coefficient, and interprets the magnitudes on that assumption — and, if it suits them, feeds your table into a meta-analysis.
There are two clean ways to handle it. The first is to use both triangles: Pearson below the diagonal, Spearman above, with a sentence in the note that leaves no ambiguity. The second, and the better one when only two or three variables need ranks, is a superscript letter on those variable names with a specific note: "aCoefficients involving this variable are Spearman's rs."
The symbols matter, because they are all that distinguishes one coefficient from another once the table is reproduced on its own in a slide: plain r is Pearson, Spearman is rs (or rho, set roman, if the journal prefers it) and Kendall is tau, roman as well. The choice is made before you look at the results: picking whichever coefficient yields more asterisks is a quiet form of p-hacking, and it shows. The criteria for deciding are in how to report a Spearman correlation in APA 7.
If a regression, a factor analysis or a structural equation model follows the table, the matrix you publish is the one that feeds that model, not a prettier version computed on the side. The other one, if you ran it, belongs in supplementary material as a sensitivity check, with a sentence in the text saying the pattern does not change. A reviewer who finds two different matrices from the same study with no explanation stops reading the results and starts reading the methods with a magnifying glass.
Already drafted it? Paste your Results section into Review one section: it reads it the way a reviewer would and returns the first objection, attached to the sentence that causes it.
Frequently asked questions
What goes on the diagonal of an APA correlation table?
By default a dash, or nothing at all: an empty cell. Writing 1.00 adds no information, because a variable correlates with itself by definition. Many organisational and health psychology journals ask for the reliability coefficients there in brackets — (.89), (.84) — which is perfectly acceptable as long as the table note declares it.
Do you put Cronbach's alpha on the diagonal of a correlation table?
You can. It is a widespread convention and APA 7 does not forbid it: the alpha goes in brackets in each variable's own cell, and the general note declares it ("Cronbach's alpha coefficients appear in parentheses along the diagonal"). The cleaner alternative, if some of your variables are not scales, is an alpha column next to M and SD. Whichever you pick, keep it identical across every table in the manuscript.
Should an APA correlation matrix use the upper or the lower triangle?
The lower one, by convention. The matrix is symmetric and the upper triangle repeats exactly the same coefficients, so printing both means printing every number twice. Use the upper half only when it carries different information — another group, another time point, another coefficient — and then the note must state what each half contains.
Where do means and standard deviations go in a correlation table?
Usually M and SD are the first two columns, between the variable name and the matrix. Putting them as the last two rows under the matrix is also correct, and it is what people do when there are so many variables that columns run short. In either position they keep their leading zero (SD = 0.97), and it is worth adding the possible range of each scale to the note.
How do you show significance in an APA correlation table?
With the probability note on the last line under the table: *p < .05. **p < .01. ***p < .001. No space between the asterisk and the p, the p italicized, no leading zeros, and only the levels the table uses. A growing number of journals prefer the exact p in brackets in each cell and no asterisks, because a star destroys information: p = .041 and p = .012 both get exactly one.
Can you put Pearson and Spearman correlations in the same table?
Yes, provided the reader can tell them apart. The two accepted solutions are giving each coefficient its own triangle (Pearson below the diagonal, Spearman above) or flagging the rank-based variables with a superscript letter explained in a specific note. The symbols help: r for Pearson, rs for Spearman. What does not work is mixing them silently.
What if the table is perfect and the text says something else?
An immaculate table with a coefficient that does not match the paragraph discussing it is a rejection in good typography. Previum reads the whole manuscript and checks what the tables say against what the text says, before an editor does it for you.
Run your manuscript through PreviumThe correlation table is the first thing a reviewer looks at and the only part that gets reproduced whole when somebody puts your study on a slide, so it earns the half hour it takes to get right. If you are closing the results section of a thesis or a paper and would rather have someone go through it with the data in front of them, our statistical consultancy works through it with you, coefficient by coefficient, with the target journal on the table.