Dixon's Q Calculator — Instant Outlier Check
Enter your dataset as comma-separated numbers. The calculator sorts, computes Q for the most extreme value, and compares it to Qcrit at your chosen confidence level.
What Is Dixon's Q Test?
Dixon's Q Test is a statistical procedure for deciding whether the most extreme value in a small dataset — either the highest or the lowest — is a genuine outlier or a plausible result from sampling variation. It was introduced by W. J. Dixon in 1950 and remains the standard outlier test in analytical chemistry, pharmaceutical quality control, and environmental laboratory work.
The test works by computing a ratio called the Q statistic. This ratio compares the "gap" between the suspected value and its nearest neighbour against the full "range" of the dataset. A large Q means the suspected value sits far from the rest of the data relative to how spread out the data is overall. If Q exceeds the critical value from this table, the suspected value is rejected at the chosen confidence level.
Key limitation: Dixon's Q Test tests one suspected outlier at a time. Applying it repeatedly on the same dataset to remove multiple values inflates the false-positive rate. For datasets with two or more potential outliers, use the generalized ESD test. For n > 30, Grubbs' Test is more appropriate.
Dixon's Q Table — Critical Values (n = 3 to 30)
Select the confidence level matching your analysis. If Qcalc > Qcrit at your chosen level, the suspected outlier is statistically significant and may be rejected. Click any row to load the critical value into the calculator panel.
Values are Qcrit for the standard r₁₀ form: Q = (x₂ − x₁)/(xₙ − x₁) or Q = (xₙ − xₙ₋₁)/(xₙ − x₁). Source: Dixon, W. J. (1950) and Rorabacher, D. B. (1991). Reject the suspected outlier if Qcalc > Qcrit.
The Dixon Q Formula
Sort your dataset into ascending order so that x₁ is the smallest value and xₙ is the largest. The Q statistic depends on which end of the dataset holds the suspected outlier.
Testing the lowest value (suspected x₁ is too small)
x₂ is the second-smallest value. xₙ is the largest value. xₙ − x₁ is the full data range.
Testing the highest value (suspected xₙ is too large)
xₙ₋₁ is the second-largest value. xₙ − x₁ is the full data range.
Decision rule
If Qcalc ≤ Qcrit → Retain the value (not a statistically significant outlier)
The Sort–Calculate–Compare–Decide Framework
Every Dixon Q Test follows the same four-step procedure. This framework covers the full decision pathway from raw measurements to a documented outlier decision.
Worked Example 1 — Analytical Chemistry: Titration Results
Scenario: A laboratory technician performs seven titrations and records the following volumes in mL: 20.14, 20.17, 20.19, 20.22, 20.24, 20.26, 20.51. The value 20.51 looks suspiciously high. Should it be rejected at 95% confidence?
Solution — Step by Step
| Step | Action | Result |
|---|---|---|
| 1 | Sort data ascending | 20.14, 20.17, 20.19, 20.22, 20.24, 20.26, 20.51 |
| 2 | Identify suspect (highest) | xₙ = 20.51, xₙ₋₁ = 20.26, x₁ = 20.14 |
| 3 | Calculate Q | Q = (20.51 − 20.26) / (20.51 − 20.14) = 0.25 / 0.37 = 0.676 |
| 4 | Look up Qcrit | n = 7, 95% confidence → Qcrit = 0.507 |
| 5 | Decision | 0.676 > 0.507 → Reject 20.51 as an outlier |
Interpretation
At 95% confidence, the value 20.51 mL is a statistically significant outlier and may be excluded from the mean calculation. The remaining six values give a mean of 20.20 mL (SD = 0.042 mL), compared to 20.25 mL if the outlier is included. The excluded value is likely the result of an instrument error, an incomplete titration, or a data transcription mistake — document the reason in the laboratory record.
Worked Example 2 — Environmental Lab: Water Quality Measurement
Scenario: An environmental technician tests nitrate concentration (mg/L) from five replicate water samples: 3.2, 3.5, 3.6, 3.7, 5.1. The value 5.1 appears anomalously high. Test at 95% confidence.
Solution
Qcalc = (5.1 − 3.7) / (5.1 − 3.2) = 1.4 / 1.9 = 0.737
Qcrit at n = 5, 95% confidence = 0.565
0.737 > 0.565 → Reject 5.1 mg/L as an outlier
The anomalous reading is likely contamination of a single sample vessel or instrument calibration drift. The decision to reject should be accompanied by an investigation into the cause. Mean without outlier = 3.50 mg/L; mean with outlier = 4.02 mg/L — a 15% difference that matters in regulatory reporting contexts.
Choosing a Confidence Level: 90%, 95%, or 99%?
The three available confidence levels represent three different trade-offs between the risk of incorrectly rejecting a legitimate value (Type I error) and the risk of retaining a genuine outlier (Type II error).
90% Confidence (α = 0.10)
The most permissive threshold. There is a 10% chance of rejecting a value that belongs to the dataset. Use when the cost of missing an outlier is higher than the cost of discarding a legitimate data point — for example, in exploratory screening.
95% Confidence (α = 0.05) — Standard
The standard choice for most laboratory and chemistry applications. A 5% chance of a false rejection is accepted as a reasonable balance. ASTM E178 and many ISO laboratory standards use this level as the default for outlier testing.
99% Confidence (α = 0.01)
The most conservative threshold. Only extreme values are rejected — the test requires very strong evidence. Use for regulatory submissions, clinical laboratory data, or whenever the consequences of incorrectly removing a valid measurement are serious.
Practical guidance: Decide on your confidence level before collecting data, not after inspecting results. Choosing the level that happens to make a specific value significant is a form of data manipulation and violates good statistical practice. See the discussion of significance levels on Statistics Fundamentals.
Dixon's Q Test vs Grubbs' Test: When to Use Each
Two tests dominate single-outlier detection in laboratory science. The right choice depends on your sample size and whether you are working to a specific standard.
| Feature | Dixon's Q Test | Grubbs' Test |
|---|---|---|
| Ideal sample size | n = 3 to 7 | n = 6 to 25 |
| Table range covered | n = 3 to 30 | n = 3 to 100+ |
| Calculation complexity | Very simple — ratio only | Requires mean and SD |
| Normality assumption | Moderate (robust at small n) | Stronger assumption of normality |
| Multiple outliers | Not supported (one at a time) | Not supported (one at a time) |
| Common standards | ASTM E178, ISO 5725 | ASTM E178, NIST/SEMATECH |
| Masking risk | Low at very small n | Higher when two outliers exist |
Bottom line: For the smallest datasets (n = 3 to 7) common in analytical chemistry — triplicate measurements, calibration checks, or pilot experiments — Dixon's Q Test is the practical choice. As sample sizes grow beyond 10, Grubbs' Test generally produces more reliable decisions. When your laboratory procedure specifies a test method, follow that specification regardless of personal preference.
Dixon's Q Test Across Laboratory Disciplines
The test applies in any field where replicate measurements are taken and a single unusual result raises concern.
Pharmaceutical QC: Tablet Dissolution Testing
USP dissolution tests require six-unit measurements per batch. A single tablet that dissolves far slower or faster than the others triggers a Dixon Q evaluation before the batch fails specification. Because the test is confirmed with n = 6, it falls squarely in the zone of maximum reliability for Dixon's Q Test.
Environmental Monitoring: Air Quality Sampling
Regulatory air quality monitors take multiple readings per sampling period. When one reading is substantially higher than the others — possibly due to a passing vehicle or brief equipment malfunction — Dixon's Q Test provides a statistically defensible basis for investigating or flagging that reading before reporting to regulators.
Materials Testing: Tensile Strength Measurements
Engineering labs testing material specimens often work with five to ten samples. A specimen that fractures at an unusually low stress — indicating a pre-existing flaw rather than bulk material properties — can be evaluated with Dixon's Q Test to decide whether to exclude the value from the reported specification.
Food Safety: Microbiological Counts
Microbiology laboratories plating replicate samples from the same food batch may occasionally produce one plate with an anomalous colony count due to pipetting error, contamination, or miscounting. Dixon's Q Test at 95% confidence helps distinguish a genuine contamination event from a measurement artifact.
Symbol and Concept Glossary
Every term used in a Dixon Q Test interpretation is defined below, with the most common point of confusion noted for each.
| Symbol / Term | Definition | Common Error |
|---|---|---|
| n | Total number of values in the dataset | Forgetting to include the suspected outlier when counting n |
| x₁, xₙ | Smallest (x₁) and largest (xₙ) values after sorting | Using unsorted positions rather than sorted order |
| Gap | Distance between the suspected outlier and its nearest neighbour | Measuring to the wrong neighbour (second-nearest instead of nearest) |
| Range | xₙ − x₁: the full spread of the sorted dataset | Using standard deviation instead of range — wrong formula |
| Qcalc | The calculated Q statistic = gap / range; always between 0 and 1 | Computing range as second-largest minus second-smallest |
| Qcrit | The critical value from this table for n and chosen confidence level | Using the wrong confidence level row or confusing α with confidence % |
| α | Significance level; probability of rejecting a legitimate value. α = 1 − confidence | Confusing α = 0.05 with 95% — they are related as confidence = 1 − α |
| Outlier | A value that is statistically inconsistent with the rest of the dataset at the chosen confidence level | Treating statistical significance as sufficient justification to discard — always investigate the cause |
Assumptions, Limitations, and Best Practices
Dixon's Q Test works well when used correctly. These conditions determine when it is and is not appropriate.
When to use Dixon's Q Test
- Sample size n = 3 to 30 (most reliable: 3 to 10)
- Data are approximately normally distributed
- Only one suspected outlier at a time
- Outlier is at the minimum or maximum of the sorted dataset
- Replicate measurements from a controlled process
When not to use Dixon's Q Test
- Sample size n > 30 (Grubbs' Test or ESD preferred)
- Two or more outliers are suspected simultaneously
- Data are strongly skewed or non-normal
- Goal is exploratory analysis, not confirmatory testing
- Applying repeatedly to the same dataset
Statistical rejection ≠ scientific dismissal
A statistically significant Q tells you a value is unlikely to belong to the same distribution as the rest of your data. It does not tell you why. Before removing any data point, investigate the probable cause — instrument malfunction, transcription error, sample contamination, or an unexpected real phenomenon. Documenting the reason is required by most laboratory quality management systems, including ISO/IEC 17025.
Frequently Asked Questions About Dixon's Q Test
What is Dixon's Q Table?
Dixon's Q Table lists the critical values Qcrit for the Dixon Q statistic at three confidence levels — 90%, 95%, and 99% — for sample sizes n = 3 to 30. If your computed Q exceeds Qcrit at n and your chosen confidence level, the suspected outlier is statistically significant and may be rejected from the dataset.
How do you find Q critical in Dixon's Q Table?
Count the number of values in your complete dataset (including the suspected outlier). That count is n. Select the confidence level tab — typically 95% for laboratory use. Find the row where n matches your sample size. The number in that row is Qcrit. If Qcalc > Qcrit, reject the outlier.
What happens if Q is less than Q critical?
If Qcalc ≤ Qcrit, you do not have statistical grounds to reject the suspected value at the chosen confidence level. Retain it in your dataset. A low Q does not prove the value is correct — it means the evidence is insufficient to conclude it comes from a different distribution than the rest of your data.
Can I run Dixon's Q Test more than once on the same dataset?
Not reliably. Applying Dixon's Q Test repeatedly on the same data — removing one outlier and then testing the next — is sometimes called sequential outlier testing or "cleaning" a dataset. Each additional test inflates the overall Type I error rate above the stated confidence level. If you suspect more than one outlier, use the generalized ESD test, which is designed for multiple simultaneous outlier testing while controlling the error rate.
What if my sample size is larger than 30?
Dixon's Q Test was designed specifically for small samples. For n > 30, Grubbs' Test has better statistical properties, or you can use the Z-score method with the hypothesis testing framework. The NIST/SEMATECH Engineering Statistics Handbook covers outlier detection methods for larger datasets in detail.
Is Dixon's Q Test the same as the Q test in chemistry?
Yes. What chemists call "the Q test" or "the Q rejection test" in analytical chemistry textbooks is Dixon's Q Test, named after W. J. Dixon who derived the critical values in 1950. It is the same procedure — same formula, same table, same decision rule — just referred to differently across disciplines.
How does confidence level affect the decision?
A higher confidence level means a higher Qcrit, which makes it harder to reject a value. At 99% confidence, you are accepting only a 1% risk of wrongly discarding a legitimate data point — so the bar for rejection is much higher than at 90%. For example, at n = 7, Qcrit = 0.434 (90%), 0.507 (95%), and 0.637 (99%). A value with Qcalc = 0.55 would be rejected at 90% and 95% but retained at 99%.
Sources & Further Reading
The critical values in this table are taken from the sources below, which established and validated the standard Qcrit values for Dixon's Q Test.
Dixon, W. J. (1950). Analysis of extreme values. The Annals of Mathematical Statistics, 21(4), 488–506. doi:10.1214/aoms/1177729747 — The original paper introducing the Q statistic and its critical value distributions.
Rorabacher, D. B. (1991). Statistical treatment for rejection of deviant values: Critical values of Dixon's Q parameter and related subrange ratios at the 95% confidence level. Analytical Chemistry, 63(2), 139–146. doi:10.1021/ac00002a010 — The most-cited reference for Dixon Q critical values in analytical chemistry, extending the original tables.
NIST/SEMATECH e-Handbook of Statistical Methods (2013). Section 7.1.6: Detection of Outliers. National Institute of Standards and Technology. itl.nist.gov — Comprehensive U.S. government reference for laboratory outlier detection, covering both Dixon and Grubbs methods.
ASTM E178-21. Standard Practice for Dealing With Outlying Observations. ASTM International. astm.org — The regulatory standard specifying when and how to apply Dixon's Q Test and Grubbs' Test in engineering and laboratory contexts.
ISO 5725-2:2019. Accuracy (Trueness and Precision) of Measurement Methods and Results — Part 2. International Organization for Standardization. iso.org — International laboratory standard covering outlier identification in interlaboratory comparisons and measurement uncertainty.
Related Statistical Tables & Resources
Understanding What Dixon's Q Table Tells You
Why Qcrit Decreases as n Increases
With more data points, the range is more stable and representative of the true spread. A value that sits far from its nearest neighbour in a dataset of three could simply be sampling variation — but that same gap in a dataset of 20 becomes genuinely implausible. Larger n means you can detect outliers with less extreme Q values, so Qcrit falls as n rises.
The Relationship Between Q and the Data Range
Because Q divides by the full data range, it is a dimensionless relative measure. Two datasets with completely different measurement units can produce the same Q. This scaling makes the critical value table universal — the same Qcrit applies whether your measurements are in milligrams, millilitres, metres per second, or parts per billion.
Why n = 3 Has a Very High Qcrit
At n = 3, Qcrit is 0.970 at 95% confidence — almost 1. With only three measurements, any one value is necessarily close to being the range maximum divided by the full gap. Statistical power is extremely low. This is why triplicate measurements are the minimum accepted for Dixon Q testing, and replicating at n = 5 or 6 is strongly preferred in analytical work. See the outliers guide on Statistics Fundamentals for more detail.