APA 7 Italics, Decimals and Leading Zeros: The Complete Rule

This is the question that arrives once the analyses are done and all that is left is writing them up: is Cohen’s d italicized? What about Cronbach’s alpha? Does .55 take a zero in front of it or not? There are three rules here, not thirty, and once you have them they apply themselves to any statistic you meet, including one you have never seen before. This page gives you all three: what goes in italics and what does not, the real leading-zero rule (which is not the one most people repeat), and how many decimals each figure takes. If what you need are ready-made sentences instead, they are in how to report results in APA 7.

The italics rule fits in one sentence

Here it is: a statistical symbol or a variable written with a Latin letter is italicized, and the same kind of symbol written with a Greek letter is not. Everything else in this article is a special case of that sentence. There is no list to memorize. You look at which alphabet the letter comes from and you have your answer.

Italicized. Descriptives (M, Mdn, SD, SE), sample sizes (N, n), degrees of freedom (df), test statistics (t, F, z, U, W, H, Q), the p value, Latin-letter association indices (r, R, R²), Latin-letter effect sizes (Cohen’s d, Hedges’ g, Cohen’s f), unstandardised regression coefficients (B or b), and any algebraic variable you define yourself, such as X, Y or k.

Not italicized. Every Greek letter, with no exceptions: Cronbach’s alpha (α = .86), McDonald’s omega (ω = .88), the standardized regression coefficient (β = .29), chi-square (χ²), eta squared (η² = .06), phi, Kendall’s tau, Spearman’s rho when you write it as ρ, kappa (κ = .71) and Glass’s delta. Abbreviations that name something rather than carry a value stay upright too: CI, log, ln, exp, min, max, ANOVA, and SEM when it means structural equation modeling rather than the standard error of the mean. So do subscript labels, which get their own section below.

The pair that goes wrong most often is the regression pair, and it is also the most visible, because both coefficients sit in the same table two columns apart. The unstandardised one is B, a Latin letter, italicized; the standardized one is β, a Greek letter, upright. The full line reads B = 0.42, SE = 0.13, β = .29, t(196) = 3.23, p = .001. Setting β in italics, or B without them, are the two commonest typographic errors in a regression table, and a copy-editor fixes them one at a time.

Subscripts, superscripts and the scaffolding that stays upright

Italic and upright type can live inside a single symbol, and the boundary is simple: a subscript that is a label stays upright, and a subscript that is a variable is italicized. In M_diff the M slants and “diff” does not, because “diff” abbreviates the word difference and stands for no number. The same goes for “adj” in R²_adj. In X_i, by contrast, the i is italicized, because it is an index running over values.

That one rule settles the symbols you meet most often in psychometrics. Partial eta squared is η²_p: eta upright because it is Greek, the 2 upright because it is an exponent, and the subscript p upright as well, because there it abbreviates the word partial rather than naming the p value. It is the only p in your whole manuscript that is not italicized. Omega hierarchical is ω_h with an upright h, and Spearman’s correlation written the Latin way is r_s, with the r slanted and the s upright.

What never takes italics is the scaffolding: the brackets, the degrees of freedom inside them, the equals sign, the digits and the exponent. You write t(118) = 3.02, p = .003, d = 0.55, and only the t, the p and the d slant. It sounds like a triviality until you open a results section where whole paragraphs lean over because somebody selected the sentence and hit the italics button.

Two symbols deserve naming, because they are mis-set in opposite directions. Chi-square carries no italics of its own: χ²(1, N = 312) = 6.24, p = .012 — chi is Greek and the 2 is an exponent, so the only slanted characters on that line are the N and the p. R² works the other way round: the R is Latin and italicized, the 2 is not. And if your manuscript carries matrix algebra, vectors and matrices are set in bold, which is a third category rather than a variant of italics.

Leading zeros: the rule is “can exceed 1”, not “is small”

APA does not say “drop the zero when the number is small”. It says use a leading zero when the statistic can exceed 1 in absolute value, and drop it when the statistic cannot. That distinction resolves half the confusion on its own, because the criterion is set by the statistic, not by the value you happen to have obtained. Cohen’s d can perfectly well come out at 2.30, so it always takes a leading zero, including when it lands below one: write d = 0.55, never d = .55. It is the formatting error I correct most often in a results section.

No leading zero. The p value (p = .003), correlations (r = .28, ρ = .24, φ = .17), R² and adjusted R² (R² = .21), the standardized beta (β = .29), reliability and agreement coefficients (α = .86, ω = .88, ICC = .84, κ = .71), proportions of variance (η² = .06, ω² = .05), Cramér’s V, plain proportions, and by convention the fit indices of a model (CFI = .96, RMSEA = .05, SRMR = .04). They all share the same property: they are bounded and cannot climb past 1. The correlation case, with its full table and its intervals, is worked through in how to report a correlation in APA 7.

Leading zero. Means and standard deviations when the scale allows it (M = 0.94, SD = 0.38), unstandardised regression coefficients and their standard errors (B = 0.42, SE = 0.13), test statistics (t = 3.02, F = 4.82, z = 2.85), the d family of effect sizes (d = 0.55, g = 0.31), odds ratios (OR = 1.84), ages, and any score still on its original metric. If the number could pass one, the zero stays.

Two details slip past almost everybody. First, “in absolute value” means a negative d is written d = −0.55, with its zero and with a real minus sign rather than a hyphen. Second, a confidence interval inherits the format of whatever it accompanies, so you write d = 0.55, 95% CI [0.19, 0.92] for an effect size and r = .28, 95% CI [.14, .41] for a correlation — with the same number of decimals as the point estimate. If you want the substance behind those numbers rather than their typography, it is in how to interpret Cohen’s d.

How many decimals each figure takes

The APA 7 principle is to round as far as you can without losing useful precision, and in practice that means two decimals for nearly everything: test statistics, correlations, effect sizes, means, standard deviations and regression coefficients. Four decimals do not make a finding more rigorous. They make the table harder to read and they tell the reviewer you pasted the software output without looking at it.

The p value is the exception, at two or three decimals and with a floor. When p falls below .001 you do not print the number; you print p < .001. And there is one figure that simply does not exist: p = .000. That is not a p value, it is what SPSS shows when it rounds something smaller than .0005 to three decimals. Leaving it in the manuscript asserts that the probability is exactly zero, which it cannot be. Pick two decimals or three and hold to it across the paper: p = .03 and p = .031 are both acceptable, mixing them in one paragraph is not.

Degrees of freedom are whole numbers and take no decimals — unless they have been corrected. Welch’s t test and the Greenhouse-Geisser correction both produce fractional degrees of freedom, and those are reported to two decimals: t(94.61) = 2.28, p = .025, or F(1.71, 85.50) = 7.36, p = .002, η²_p = .13. Seeing the decimals tells your reader you applied the correction without your having to add a sentence saying so. N and n are always whole numbers. Percentages take one decimal or none, and with small samples none is better: in a group of 40 people each participant is worth 2.5 percentage points, so writing 32.5% claims a precision the data do not have.

Means and standard deviations are rounded to what the instrument measures, not to what the software prints. A 1-to-5 Likert item supports two decimals (M = 3.18, SD = 0.74); a reaction time in milliseconds needs none; an age needs one at most. Within a single table column every value carries the same number of decimals and aligns on the point. And rounding happens at the end, never halfway: if you round a mean to two decimals and then compute d from the rounded figure, the d comes out wrong. To check a whole sentence at once — italics, decimals and zeros together — the APA 7 formatter hands it back corrected.

The questions people actually send: Cohen’s d, p, alpha and N

Cohen’s d. Italicized, and with a leading zero: d = 0.55. It satisfies both rules at once, because it is a Latin letter and because it is a statistic that can pass one. Hedges’ g behaves identically (g = 0.31), as does Cohen’s f. The exception in the family is Glass’s delta: Δ is Greek, so it stays upright even though it is the same kind of index, a standardized mean difference that happens to be divided by the control group’s standard deviation alone.

The p value. Always italicized, everywhere: in running text, inside a table cell, in a figure note, and in the line explaining what the asterisks mean. Always without a leading zero, because a probability cannot exceed one. The only upright p in the whole manuscript is the subscript of η²_p.

Cronbach’s alpha. Upright and with no zero: α = .86. This is the question that arrives by email most often, and the answer surprises people because it runs the opposite way to Cohen’s d — no italics and no zero. Both halves come from the same two rules: alpha is a Greek letter, and alpha is bounded above. McDonald’s omega behaves the same way (ω = .88). And if you spell the name out instead of using the symbol — “Cronbach’s alpha was .86” — there are still no italics, because ordinary words do not slant.

N and n. Both italicized, and the difference between them is not decorative: the capital is the total sample and the lower case is a subgroup. N = 184 participants, of whom n = 61 were in the control condition. When a reviewer sees n = 184 in a table and N = 184 in the text, they cannot tell whether participants have gone missing. One more pair that travels together but behaves differently: OR is a statistic and takes italics, while CI names an interval rather than a value and stays upright. OR = 1.84, 95% CI [1.21, 2.80], p = .004.

Where it breaks: SPSS, Word, Excel and LaTeX

SPSS output carries no italics at all. None, in any table. Paste the coefficients table into your manuscript and every symbol arrives upright, so the italics have to go in by hand afterwards. The Sig. column is the other trap: it rounds to three decimals and prints .000, which in your write-up is p < .001. R and jamovi are no better — summary() gives you plain text, and jamovi tables paste as images or as unformatted cells.

Word loses italics on paste, and spreadsheets ignore the zero rule entirely. Choose “keep text only” or “merge formatting” in Word and your t and your p quietly straighten up, and you never look at them again; find and replace will not bring them back either, because you cannot ask Word to italicize every p without also slanting every word that contains one. Excel and Google Sheets add the mirror-image problem: a correlation of .28 is stored as a number and printed as 0.28, so a table built in a spreadsheet arrives in the manuscript with a zero in front of every correlation, alpha and beta in it. Both have the same remedy, and it is one read-through of the results section and its tables with your eye on nothing else, done last, once you have stopped pasting things in.

LaTeX has the opposite problem. Math mode slants lowercase Greek by default, so $\chi^2$ and $\alpha$ come out italicized exactly where APA wants them upright. If the journal is strict about typography you need the upright variant (the upgreek package, for instance); if it is not, plenty of journals accept the engine’s slanted Greek and never flag it. What does not hold up is half a paper set one way and half the other.

Tables and figure notes follow the same rules as the text, and they are where the rules get lost, because they are built separately and often rebuilt the night before submission. A column heading reads M, not M; the note underneath reads N = 184 and “* p < .05” with the p slanted; and every value in a column carries the same number of decimals. To see all of this applied across a full section, there is a complete worked one in the APA 7 results section example.

Already drafted it? Paste your Results section into Review one section and you get the first objection a reviewer would raise, quoting the exact sentence from your text. Free, no sign-up.

Frequently asked questions

Is Cohen’s d italicized in APA 7?

Yes, and it also takes a leading zero: d = 0.55. It is italicized because it is a Latin letter standing for a statistic, and it takes the zero because d can exceed 1 in absolute value, even if your own study never gets there. Hedges’ g follows the same pattern. Glass’s delta is the exception: it is a Greek letter, so it stays upright.

When do you use a leading zero in APA style?

When the statistic can exceed 1 in absolute value. The rule is set by the statistic, not by the number you obtained. Means, standard deviations, unstandardised B coefficients, t, F, z, Cohen’s d and odds ratios all keep the zero. The p value, r, R squared, standardized beta, alpha, omega, kappa, eta squared and proportions all drop it, because they are bounded by 1.

Is the p value italicized in APA 7?

Yes, always and everywhere: in the text, inside tables, in figure notes and in the line explaining the asterisks. It takes no leading zero, it is reported to two or three decimals, and once it falls under .001 it is written as less than .001 rather than as a figure. The only upright p in the manuscript is the subscript of partial eta squared, where p abbreviates the word partial.

Is Cronbach’s alpha italicized?

No. Alpha is a Greek letter and Greek letters stay upright: α = .86. It takes no leading zero either, because alpha is bounded above and cannot exceed 1. McDonald’s omega, eta squared, standardized beta, kappa and chi-square all stay upright too, although chi-square does keep a leading zero when it lands below one, because it is not bounded: χ² = 0.42. If you spell the name out in words instead of using the symbol, there are still no italics.

How many decimal places should a p value have in APA 7?

Two or three, applied consistently across the paper. Below .001 you do not give the figure at all; you report it as less than .001. And p = .000 is not a valid value, it is what SPSS prints when it rounds something smaller than .0005 to three decimals. Everything else takes two decimals, except degrees of freedom, which are whole numbers unless Welch or Greenhouse-Geisser has corrected them.

Is N italicized, and when is it capital N or lowercase n?

Both are italicized. Capital N is the total sample for the study and lowercase n is a subgroup: N = 184 participants, of whom n = 61 were in the control condition. Both are whole numbers and take no decimals. Mixing them up leaves the reader unable to tell whether a table describes the whole study or a single condition.

Three rules and you are done: Latin letters italicized and Greek letters upright, a leading zero only when the statistic can pass 1, and two decimals unless the p value asks for three. That covers the formatting of any figure you will ever produce, including one you have not met before. If the hard part is not the typography but the analysis underneath it, we can work through that together in my statistical consulting.