KURT in Google Sheets: Sample Excess Kurtosis Explained
Use KURT in Google Sheets to compare tail behavior, distinguish corrected excess kurtosis from raw fourth moments and check reproducible datasets independently.
Verification status: See the evidence and test scope described in this guide. How formula examples are checked
Quick answer: if A2:A6 contains 2, 4, 6, 8 and 10, =KURT(A2:A6) has expected result -1.2. The negative value is meaningful; kurtosis is not a probability and need not lie between 0 and 1.
Syntax and the convention behind the result
=KURT(value1, [value2, ...])
The Google KURT reference accepts values or ranges and permits additional arguments. Google says text is ignored and that Sheets supports an arbitrary number of arguments despite a nominal 30-argument specification.
The numerical convention matters. Google’s linked published examples give approximately -0.6811784575 for 3, 5, 6, 8, 10 and 0.3905325444 for 3, 5, 6, 8. Both reproduce the bias-corrected sample excess-kurtosis formula. They do not reproduce an unadjusted population fourth-moment ratio.
This convention identification is an inference from independently reproducing Google’s own published numerical examples. It is not a native execution claim or an assumption borrowed from another spreadsheet.
Worked example: change one tail observation
Enter the following two synthetic datasets:
| Row | Column A | Column B |
|---|---|---|
| 2 | 2 | 2 |
| 3 | 4 | 4 |
| 4 | 6 | 6 |
| 5 | 8 | 8 |
| 6 | 10 | 30 |
In D2 and D3, enter:
=KURT(A2:A6)
=KURT(B2:B6)
Expected results are -1.2 and 4.2532544379. Moving one observation from 10 to 30 changes the tail contribution substantially. The example illustrates sensitivity; it does not prove that the second dataset arose from any particular statistical distribution.
Do not interpret a high KURT value as an automatic instruction to delete the largest measurement. First determine whether the value is a valid observation, a recording error or a different process that should be analyzed separately.
Independently reproduce the first result
For A2:A6, n = 5 and the mean is 6. Deviations are -4, -2, 0, 2 and 4. Their squared sum is 40 and their fourth-power sum is 544. Thus the average squared deviation m2 is 8, and the average fourth power m4 is 108.8.
The unadjusted excess ratio is m4/m2^2-3, which equals -1.3. Applying the sample correction gives:
=(5-1)/((5-2)*(5-3))*((5+1)*(108.8/8^2-3)+6)
Expected result: -1.2, matching the KURT example. This calculation shows why an independently computed unadjusted ratio of -1.3 is not evidence that the spreadsheet result should also be -1.3.
NIST explains the distinction between kurtosis and excess kurtosis. Excess kurtosis shifts the normal-distribution reference from 3 to 0. Sample-size corrections introduce another difference that must be stated when comparing software or reports.
Advanced application: compare groups without pooling them
If A2:A6 and B2:B6 represent separate processes, calculate each group’s KURT separately before considering a pooled summary. A combined dataset answers a different question about the mixture. Differences in group means alone can change the shape of that mixture.
Kurtosis is dimensionless. Adding the same constant to every observation leaves it unchanged, as does multiplying every observation by the same nonzero scale factor. For example, converting the A-column values to 102, 104, 106, 108, 110 preserves the expected result -1.2. Converting them to 4, 8, 12, 16, 20 also preserves it.
These invariance checks can help detect inconsistent unit conversions. They do not make kurtosis stable when observations themselves are added, removed or selectively transformed.
Limits of the summary
Google describes kurtosis using peakedness. For interpretation, focus also on tail weight and extreme observations; a single peak-height description is not enough to characterize a distribution. A value close to zero does not establish normality, and positive excess kurtosis does not uniquely identify a model.
Small samples can produce volatile estimates. Use the raw values and other summaries alongside KURT rather than reducing an entire dataset to one shape number.
Errors and edge cases
Google documents #DIV/0! when fewer than two values are supplied. That statement does not establish that every two- or three-value input is valid. The corrected formula used in the numerical checks has denominators involving n minus 2 and n minus 3; the examples therefore use at least four varying numeric values.
Native handling of two or three observations, constant datasets, blanks and logical values was not established here. Text is documented as ignored, but a missing observation should not be converted into a numeric zero merely to avoid an error.
Availability and verification
KURT is documented in current rolling Google Sheets. Original introduction date, first release and last functioning numbered version were not established.
NOT_TESTED_IN_GOOGLE_SHEETS. Google help and its linked read-only published examples were inspected on October 17, 2026. Original outputs and the two published reference values were independently checked from central moments and numerical kurtosis calculations. No native execution or screenshot is claimed.