Choose the definition from the question, not the answer size
Population standard deviation divides the sum of squared deviations by the observation count n before taking the square root. Sample standard deviation uses n − 1 instead. The choice depends on what the list represents and which statistic the question requires, not on which answer looks more favorable. A calculator cannot identify a population or a sampling design by examining a string of numbers alone.
Suppose a hypothetical worksheet records values 3, 5, 5, 7, 10. This article uses that single list for both definitions so the effect of the divisor can be isolated. The numbers could be descriptive readings in one consistent unit; they are not a recommendation or a measured industry benchmark. Use the Standard Deviation Calculator to follow the mean, deviations and squares, keeping the selected definition alongside the reported result.
Define the population before selecting a mode
If the question describes exactly those five records as the entire collection of interest, population mode summarizes that complete entered set. If the five records are a sample used to estimate variance in a larger population under appropriate sampling assumptions, the sample equation is the relevant alternative. The same list can therefore support different questions. Merely having a small number of observations does not automatically make them a sample of something larger.
For a practical record, write a scope statement such as all five entries in this specified batch, or five selected entries from a separately defined larger collection. Those descriptions reveal more than a checkbox label. They still do not establish that the selected records are representative or independent. Sampling quality must be assessed separately, and the descriptive calculator does not turn convenience data into a probability sample by dividing by one fewer observation.
Find one mean and the signed deviations
Add the example values to obtain 3 + 5 + 5 + 7 + 10 = 30, then divide by five to get mean 6. Subtracting that mean produces deviations −3, −1, −1, 1 and 4. Their sum is zero in exact arithmetic. The mean and deviations are identical in the sample and population calculations here; changing the selected divisor does not remove a record or recalculate a different five-number list.
Square each deviation before combining them: 9, 1, 1, 1 and 16. These total 28. The table shows the original observation order so repeated fives remain separate measurements. Adding signed deviations without squaring would give zero and would not measure spread. Do not round the mean to a convenient integer in a different example merely because this particular list happens to have an exact whole-number mean.
| Observation | Value | Squared deviation from mean 6 |
|---|---|---|
| 1 | 3 | 9 |
| 2 | 5 | 1 |
| 3 | 5 | 1 |
| 4 | 7 | 1 |
| 5 | 10 | 16 |
The values total 30 and the squared deviations total 28. Both definitions use this same list and mean; only the divisor changes.
Divide by n or n minus one, then take the root
For population mode, divide the squared-deviation sum 28 by n = 5. Population variance is 5.6 and population standard deviation is √5.6, approximately 2.36643191324. For sample mode, divide the same 28 by n − 1 = 4. Sample variance is 7 and sample standard deviation is √7, approximately 2.64575131106. The calculation steps expose the denominator rather than displaying an unlabeled square-root result.
With positive observed spread and more than one value, the sample result is larger on the same list because its denominator is smaller. For this example, the ratio between the standard deviations is √(5/4), approximately 1.11803398875. The relationship is not a choice between a pessimistic and optimistic scenario. They are different definitions, and switching them simply to increase or reduce a reported result would misstate the calculation's intended scope.
Keep variance, deviation and uncertainty distinct
Variance uses squared units, while standard deviation returns to the original units. If the example readings were measured in centimeters, a population variance of 5.6 would be in square centimeters and its standard deviation would be about 2.3664 centimeters. Labeling both as centimeters would erase a meaningful distinction. A comparison also needs the same measurement scale; converting units changes the numerical values of both statistics in different ways.
Neither statistic is standard error, a confidence interval or a test of a distribution's shape. Dividing by n − 1 addresses variance estimation under the usual sampling assumptions; taking its square root does not make sample standard deviation exactly unbiased in general. The tool does not calculate an inferential uncertainty statement or decide that a reading is an outlier. Keep those separate analytical questions out of a simple descriptive result label.
Test singletons and constant lists before trusting a workflow
A one-member population containing only 9 has mean 9 and zero deviation, so population variance and standard deviation are zero. The sample equation cannot divide by n − 1 in that case, because its denominator is zero. Sample mode therefore reports an error rather than inventing a zero estimate. An empty list has no mean and is invalid in both modes, regardless of how a spreadsheet represents blank cells.
For 9, 9, 9, both definitions return zero because every observation equals the mean and both denominators are valid. Negative original values also work normally; it is their differences from the mean that are squared. A valid zero must remain an observation. If an invalid token were silently dropped, the denominator and mean could both change, creating a different data set whose plausible-looking result would not answer the original question.
Review precision and input conventions
The interactive calculator validates 1–1,000 raw observations within ±1e12, using dot decimals and optional scientific notation. Commas, semicolons, whitespace and new lines separate numbers; commas do not mark thousands. Invalid tokens and empty comma/semicolon items halt the result. It does not accept a frequency table as pairs of values and counts, choose an outlier-removal rule or fill missing readings with an automatically selected average.
Floating-point arithmetic also has limits. Closely spaced values around a large common offset can lose detail if variance is computed by subtracting large raw second moments. This implementation uses compensated sums and a corrected two-pass centered calculation instead. Inputs still become double-precision numbers, so distinctions below that resolution cannot be recovered. Nonconstant spread that underflows to zero is rejected; rescaling or specialized numerical software may be needed for extreme inputs.
Report enough information to reproduce the statistic
For the worked list, a complete note could say five observations, mean 6, squared-deviation sum 28, sample divisor 4, sample variance 7 and sample standard deviation approximately 2.64575 in the original units. A population note would instead state divisor 5, variance 5.6 and standard deviation approximately 2.36643. Those intermediate quantities make a disagreement easier to trace than an isolated rounded number labeled SD.
Preserve the raw list, units, scope statement and selected equation with the result. Use full intermediate precision and round the final displayed value according to the purpose of the report, without implying more measurement accuracy than the source supports. This educational workflow describes entered data; it does not certify a sampling method, establish practical tolerances or provide medical, engineering or financial approval. Verify the question's intended population before choosing the denominator.