Answer
Variance vs standard deviation: which one should you report?
The short answer
They carry the same information: the SD is the square root of the variance. The SD is in the units of your data, so it is the one you read and report. The variance is in squared units, which makes it hard to interpret but easy to work with, because variances of independent quantities add up. Report the SD next to the mean; use the variance inside calculations such as ANOVA, regression and pooling.
Same information, different units
The variance is the average squared distance of the values from their mean (with n - 1 in the denominator for a sample). The standard deviation is its square root. Knowing one tells you the other exactly, so neither contains information the other lacks.
What differs is the unit. If reaction times are measured in milliseconds, their variance is in squared milliseconds, a unit nobody can picture. Taking the square root brings the spread back to milliseconds, the same scale as the mean and the raw values. That is why a sentence such as "times averaged 420 ms with an SD of 60 ms" makes sense to a reader, and the same sentence with a variance of 3,600 ms² does not.
For roughly bell-shaped data the SD also has a handy reading: about two thirds of values lie within one SD of the mean and about 95% within two. No such rule of thumb exists for the variance. (For skewed data these percentages do not hold, and the SD can be a misleading summary on its own.)
Why statisticians keep the variance around
The variance earns its place in calculations. Its key property is that variances of independent quantities add: if a total is the sum of two independent parts, the variance of the total is the sum of their variances. Standard deviations do not add; they combine like the sides of a right triangle, as the square root of the sum of squares.
This additivity is what makes ANOVA (splitting total variation into between-group and within-group parts), the standard error of a mean, pooled estimates in a t test, and R-squared (the share of variance explained) work. Formulas are written in variances and the result is converted back to an SD, or a standard error, at the end. The reasons squaring was chosen in the first place are covered on the page about why the standard deviation squares the differences.
See it in R and Python
The code simulates a delivery made of two independent stages, 20 minutes on average with an SD of 3 minutes and 30 minutes with an SD of 4 minutes. It shows that the variances add, that the SDs do not, how each one changes when minutes are converted to hours, and finally summarises a small sample of 25 exam scores the way you would report it.
R
set.seed(35123)
n <- 100000
# Two independent stages of a delivery, in minutes
prep <- rnorm(n, mean = 20, sd = 3)
drive <- rnorm(n, mean = 30, sd = 4)
total <- prep + drive
# Variances add up for independent parts (9 + 16 = 25)...
round(c(var_prep = var(prep), var_drive = var(drive), var_total = var(total)), 2)
# ...standard deviations do not (3 + 4 is not 5)
round(c(sd_prep = sd(prep), sd_drive = sd(drive), sd_total = sd(total),
sum_of_sds = sd(prep) + sd(drive)), 2)
# Changing units: SD scales by the factor, variance by its square
hours <- total / 60
round(c(sd_minutes = sd(total), sd_hours = sd(hours),
var_minutes = var(total), var_hours = var(hours)), 4)
# A small sample you might report: 25 exam scores
scores <- round(rnorm(25, mean = 70, sd = 10))
round(c(n = length(scores), M = mean(scores), SD = sd(scores), variance = var(scores)), 2)Python
import numpy as np
rng = np.random.default_rng(35123)
n = 100_000
# Two independent stages of a delivery, in minutes
prep = rng.normal(20, 3, n)
drive = rng.normal(30, 4, n)
total = prep + drive
# Variances add up for independent parts (9 + 16 = 25)...
print({k: round(float(v), 2) for k, v in
{"var_prep": prep.var(ddof=1), "var_drive": drive.var(ddof=1), "var_total": total.var(ddof=1)}.items()})
# ...standard deviations do not (3 + 4 is not 5)
print({k: round(float(v), 2) for k, v in
{"sd_prep": prep.std(ddof=1), "sd_drive": drive.std(ddof=1), "sd_total": total.std(ddof=1),
"sum_of_sds": prep.std(ddof=1) + drive.std(ddof=1)}.items()})
# Changing units: SD scales by the factor, variance by its square
hours = total / 60
print({k: round(float(v), 4) for k, v in
{"sd_minutes": total.std(ddof=1), "sd_hours": hours.std(ddof=1),
"var_minutes": total.var(ddof=1), "var_hours": hours.var(ddof=1)}.items()})
# A small sample you might report: 25 exam scores
scores = np.round(rng.normal(70, 10, 25))
print({"n": len(scores), "M": round(float(scores.mean()), 2), "SD": round(float(scores.std(ddof=1)), 2),
"variance": round(float(scores.var(ddof=1)), 2)})Output from one seeded run of the R code:
var_prep var_drive var_total
9.02 16.11 25.10
sd_prep sd_drive sd_total sum_of_sds
3.00 4.01 5.01 7.02
sd_minutes sd_hours var_minutes var_hours
5.0097 0.0835 25.0971 0.0070
n M SD variance
25.00 70.44 10.21 104.17
What the numbers show
- Variances add. The stage variances were 9.02 and 16.11 and the total's variance was 25.10, matching their sum up to simulation noise (the true values are 9 + 16 = 25).
- SDs do not. The SDs were 3.00 and 4.01, which sum to 7.02, yet the total's SD was 5.01, close to the square root of 25. Adding SDs would overstate the spread of the total by 40%.
- Units behave differently. Converting minutes to hours divided the SD by 60 (5.0097 to 0.0835) but the variance by 3,600 (25.0971 to 0.0070). The SD moves with the data; the variance moves with its square.
- The reportable summary. For the 25 exam scores the run gave M = 70.44 and SD = 10.21; the variance, 104.17 squared points, is correct but tells a reader much less.
- Python agrees on the pattern. Its random numbers differ from R's, so its figures differ slightly (for example a total SD of 5.0 and a sum of SDs of 7.0), but every comparison comes out the same way.
Which one to use where
- Describing data: report the mean and SD (or the median and interquartile range for skewed data). Readers think in the data's units.
- Combining independent sources of variability: add the variances, then take the square root if you need an SD.
- Uncertainty of a mean: that is the standard error, the SD divided by the square root of n, a different quantity; see standard error vs standard deviation.
- Model output: ANOVA tables show sums of squares and mean squares, which are variance-type quantities. Mixed models report variance components; give their square roots too when you describe how much groups differ.
How to report a standard deviation in APA style (7th edition)
Report the SD next to the mean, in the same units and to the same number of decimals, and italicise the symbols. Using the exam scores from the R example:
"Exam scores averaged 70.44 points (SD = 10.21, n = 25)." Or, in parentheses: "(M = 70.44, SD = 10.21)".
Give the variance (s², or σ² for a population value) only when a reader needs it, for example variance components in a mixed model or a test of equal variances, and keep it out of descriptive sentences. In tables, a column headed SD beside the column of means is the usual layout.
Related tools and guides
- APA 7 formatter for descriptive statistics
- Why does standard deviation square the differences?
- Standard error vs standard deviation
- RMSE vs standard deviation
- Which statistical test should I use?
- Variance (Wikipedia)
More answered questions
- Where should histogram bin edges go when your data are rounded?
- Standard error vs standard deviation: what is the difference?
- Log-transforming data: when it helps, and what it changes
Working with your own data?
General answers only go so far. Send us your situation and we'll reply by email within two business days.
Ask your questionWritten by AskStats with AI assistance. This is general information, not advice for your specific data or study. When the results matter, check your approach with a qualified statistician.