Statistics are challenging! This field guide highlights statistics commonly encountered in medical literature so you can
help researchers and clinicians understand and apply them effectively (based on chapter 19 of the AMA Manual of Style, 11th edition).

Analysis of variance

Analysis of variance (ANOVA) is used to compare a dependent variable with at least 1 independent variable in a given group or sample. In this method, the F test is used to test the null hypothesis, which is that there are no differences in the means between any of the group’s samples. The t test and F statistic are the most commonly used measures of ANOVA. When the F statistic is used, degrees of freedom (df) should be included (eg, “The difference was found to be statistically significant by 1-way ANOVA (F_2,63 = 61.07; P < .001).”).

Categorical vs continuous variables

Categorical variables do not have numerical values (eg, sex, race, smoking status, education level). Continuous
variables have real number values (eg, weight, laboratory values, body temperature). Continuous data fall into 2 categories: ratios and intervals.

χ² Test

The χ² (chi-square) test determines the significance of categorical data. Use the Greek letter rather than spelling out “chi.”

Coefficient of variation

The coefficient of variation is the ratio of the SD to the mean. On first occurrence, use “coefficient of variation (s/x̄) × 100 (CV),” then “CV” thereafter.

Confidence interval

A confidence interval is the range wherein a researcher can be confident a value lies, typically with a certainty of 95%. Per AMA style, do not spell out on first instance. If no accompanying percentage is provided, query the author. CIs where both values are positive follow the form “95% CI, 1-3.5”; CIs where at least 1 value is negative follow the form “95% CI, −5 to 24.17.”

df (Degrees of Freedom)

Per AMA style, do not expand df at first occurrence. Report df as a subscript after the t test, ANOVA, χ² test, etc, statistic (eg, χ ²/₃ = 17.7, where the subscript 3 is the df).

Hazard ratio

The hazard ratio is a comparison of hazard rates between 2 groups, calculated using the Cox proportional hazards mode. Use the format “HR, 0.9,” and expand “HR” on first occurrence.

IQR vs range

The IQR (interquartile range) is the distance between the 25th and 75th percentiles, whereas the range is the highest and lowest values in a measured sample. IQRs typically accompany median values. Per AMA style, do not expand “IQR” on first occurrence, and use a hyphen, not a comma, for both IQRs and ranges (eg, IQR, 1-5).

Means

The mean is the sum of the values measured divided by the number of values. Always query the author for the SD if a mean value appears without it. Use the format “mean (SD) age, x (x) years.”

Median

The median is the midpoint in a range of values where half the values appear above and half below the variable. Always query the author for the IQR or range if a median value appears without it. Use the format “median, 3 years (IQR, 2-4 years).”

N vs n vs No.

“N” refers to the total number of units in the population being studied (eg, N = 1000), and “n” is a subgroup of that population (eg, n = 500). Use “No.” in tabular material to refer to the number of units being counted.

Odds ratio

The odds ratio is the ratio of 2 odds, such as the incidence of a disease or the relative risk of a population contracting a disease. Typically, odds ratios are accompanied by a 95% CI. Per AMA style, use the format “OR, 1.25 (95% CI, 1.10-1.40). Spell out “odds ratio (OR)” on first occurrence.

Probability

P value	Probability that a data point was achieved by chance alone (eg, if P = .02, there is a 2% chance that the value was the result of random happenstance and a 98% chance that the value is statistically significant) •• P values should never be either 1.0 or 0. Per AMA, although they may approach infinitely close to these upper and lower bounds, they can never be close enough to establish that the associated observation was absolutely predestined (P = 1.0) or absolutely impossible (P = 0) to occur because the value represents a probability •• P values ≥.01 should be expressed to 2 digits; P values <.01 should be expressed to 3 digits •• No zero before the decimal point (P = .05)
α (alpha)	Probability of a type I error—that is, a result in which sample data lead to the rejection of the null hypothesis (ie, no true difference exists between populations), despite the hypothesis actually being true in the population •• Note the roman numeral (type I, not type 1) •• A typical value is .05 •• No zero before the decimal point (α = .05)
β (beta)	Synonymous with type II error—that is, a result in which the sample data lead to the rejection of the null hypothesis, despite the hypothesis being false in the population •• Note the roman numeral (type II, not type 2) •• A typical value is .20 •• No zero before the decimal point (β = .10)

Statistics housekeeping

• In tables, use the order <variable>, <data type>, <unit of measure> (eg, Patients, No. (%) or Age, mean (SD), y). If the data type applies to an entire column, it should appear in the column header; otherwise, it should appear for each applicable row or stub.
• Use hyphens, not commas, in ranges.
• AMA style uses thin spaces, not commas, in numbers of 5 or more digits.
• Insert a thin space on either side of a mathematical operator separating a symbol from its value (eg, P < .001), but close up the operator and the value when no symbol is used (eg, aged ≥75 years).
• In mean values, do not use ± (eg, mean age, 54 ± 13 years); instead, use the format “mean (SD) age, 54 (13) years.”
• Where the word “significant” is used, query the author about the nature of the significance (eg, clinical, statistical).
• Do not use possessive eponyms (eg, Cohen d, not Cohen’s d; Student t test, not Student’s t test).

House style may deviate from AMA style, so be sure to check your client’s preferences.