Choose your reading route
eduKateSG · Why Mathematics? · Index of Dispersion, Variance-to-Mean Patterns in Count Data
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 4–7
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 8–13
Check assumptions
See where design and calibration matter.Route 4 · Chapters 14–18
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 19–26
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why is mathematics important when counts look more or less variable than a simple random-count model would suggest? The index of dispersion divides the sample variance by the sample mean. For Poisson counts with comparable exposure, variance and mean are theoretically equal, making one a useful reference value.
The index links arithmetic mean, sample variance, ratio scales and probability models. It helps students see that a benchmark is a question to investigate, not a label that diagnoses a distribution by itself.
Did you know? The index has the unit of the count scale because variance has squared units and mean has ordinary units. For pure counts the unit is often treated as a count, even though the numerical benchmark one remains central.
Full chapter index
Chapters 1–3 · Understand the question
Chapters 4–7 · Reproduce the example
Chapters 8–13 · Check assumptions
Chapters 14–18 · Explore applications
Chapters 19–26 · Study and verify
Two sequences can have the same average number of events but different clustering. A variance-to-mean ratio makes that difference visible.
It is most interpretable when observations are non-negative counts collected over comparable opportunities or exposures.
NIST defines the sample index d=s²/x-bar, where s² is sample variance and x-bar is sample mean.
NIST notes that the term “coefficient of dispersion” is also used for this ratio elsewhere, but Dataplot reserves that name for a different median-centred statistic.
| Counts | Sample mean | Squared-deviation sum |
|---|---|---|
| 1,2,3,4,5 | 3 | 10 |
Suppose five equal observation periods record 1,2,3,4 and5 events. The mean is 3.
Squared deviations from 3 are 4,1,0,1 and4, summing to10.
| Sample variance | Mean | Index |
|---|---|---|
| 10/4 = 2.5 | 3 | 2.5/3 = 0.833333 |
Sample variance divides by n−1, so s²=10/4=2.5. The index is 2.5/3=0.833333.
The arithmetic is straightforward; the interpretation still depends on exposure, independence and sampling uncertainty.
In a Poisson model, the population mean and variance are equal. That implies a population variance-to-mean ratio of one.
A sample index will fluctuate around that benchmark even when the model is appropriate.
An index below one can indicate more regular spacing or constraints than a Poisson process. Binomial-type counts can also have variance below the mean.
The worked value 0.833333 is not enough by itself to prove such a mechanism.
An index above one can arise from event clustering, changing rates, unobserved heterogeneity, dependence or mixtures. Negative-binomial and geometric count models commonly allow variance above the mean.
Several mechanisms can produce the same ratio, so investigate design and time order.
The coefficient of variation is s/x-bar. Squaring it gives s²/x-bar², which differs from s²/x-bar by another factor of the mean.
The two ratios answer different scaling questions.
The official NIST coefficient-of-dispersion definition uses average absolute deviation from the median divided by the median. It can suit positive continuous data.
Always write the formula when terminology is ambiguous.
Counts from ten minutes and one hour do not share the same opportunity. A larger count can reflect longer exposure rather than greater event intensity.
Use common intervals, analyse rates with an appropriate model, or include an exposure offset.
CHAPTER 11 OF 26 · Check assumptions
11. Rates are not automatically interchangeable
Dividing counts by exposure changes both the mean and variance scale. The familiar Poisson benchmark for raw equal-exposure counts does not transfer unchanged to arbitrary rates.
Keep the count and exposure columns available.
If every count is zero, the mean and variance are both zero and the ratio is undefined. With a very small positive mean, the ratio can be unstable.
Sparse data also make asymptotic approximations less reliable.
Autocorrelation can inflate or deflate observed variation. A burst followed by quiet periods is not the same design as independent replicated intervals.
Plot counts in sequence and consider lag structure before interpreting the ratio.
Pooling locations with different underlying rates can create overdispersion even if each location separately follows a Poisson model.
Stratification or a hierarchical model may explain the extra variance better than one pooled number.
The ratio is estimated from the same finite sample that supplies mean and variance. Small n creates substantial uncertainty.
Formal dispersion tests require assumptions; a threshold of exactly one should not be used as an automatic pass–fail rule.
Students can generate repeated sets of Poisson counts with the same mean and calculate the index for each set. The histogram will spread around one.
Then mix two different Poisson rates and observe how between-group variation increases the pooled index.
Place equal-exposure counts in one column, calculate AVERAGE and sample VAR.S, then divide variance by mean after checking the denominator.
Add a time plot and exposure column beside the calculation.
Use the sample-variance convention with degrees of freedom one. Preserve integer counts and exposure metadata.
Unit-test 1,2,3,4,5 to obtain mean3, sample variance2.5 and index0.833333.
The index does not identify a probability distribution, prove independence or correct measurement error. It is sensitive to extreme counts because variance squares deviations.
It is also unsuitable where zero is arbitrary or negative measurements occur.
Variance-to-mean reasoning appears in ecology, epidemiology, reliability, queueing, manufacturing and network monitoring. Analysts use it as an exploratory clue before choosing richer count models.
The transferable skill is connecting a summary ratio to the data-generating opportunity.
The average-deviation guide uses linear distances rather than squared ones. The Gini-mean-difference guide builds spread from pairwise gaps.
Students should verify equal exposure, show the mean and sample variance separately, and phrase conclusions as evidence rather than diagnosis.
Parents can ask whether a graph reveals clustering, trends or different groups that one ratio hides.
Calculate the index for 3,3,3,3,3 and for 0,0,0,0,15. Compare the same total arranged regularly and clustered.
Double every observation interval without adjusting the process and discuss why exposure must enter the interpretation.
### Does an index above one prove a negative-binomial model? No. Many mechanisms create overdispersion.
### Is one always the correct target? Only under a suitable equal-exposure Poisson reference model.
### Which variance denominator should I use? This article follows NIST’s sample variance convention, using n−1.
### Can the index be negative? With positive mean and ordinary variance it is non-negative; invalid scales can break interpretation.
Read the official NIST index-of-dispersion documentation and the NIST Poisson-process handbook. The worked mean, sample variance and ratio were independently recalculated.
Continue through the Mathematics Learning Hub and How Mathematics Works.
Index of dispersion turns mean and variance into a model-aware question: are these counts varying as a simple equal-rate process suggests? The answer begins with the ratio and continues with exposure, structure and evidence.
Extended practice laboratory
Practice case 1: Check count data
Setup: Use the article’s variance-to-mean example and change one declared input for case 1. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 2: Calculate the sample mean
Setup: Use the article’s variance-to-mean example and change one declared input for case 2. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 3: Calculate sample variance
Setup: Use the article’s variance-to-mean example and change one declared input for case 3. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 4: Divide variance by mean
Setup: Use the article’s variance-to-mean example and change one declared input for case 4. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 5: Compare with one
Setup: Use the article’s variance-to-mean example and change one declared input for case 5. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 6: Create equal counts
Setup: Use the article’s variance-to-mean example and change one declared input for case 6. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 7: Create clustered counts
Setup: Use the article’s variance-to-mean example and change one declared input for case 7. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 8: Create regular counts
Setup: Use the article’s variance-to-mean example and change one declared input for case 8. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 9: Change exposure time
Setup: Use the article’s variance-to-mean example and change one declared input for case 9. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 10: Use rates instead of counts
Setup: Use the article’s variance-to-mean example and change one declared input for case 10. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 11: Check zero mean
Setup: Use the article’s variance-to-mean example and change one declared input for case 11. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 12: Check meaningful zero
Setup: Use the article’s variance-to-mean example and change one declared input for case 12. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 13: Compare coefficient of variation
Setup: Use the article’s variance-to-mean example and change one declared input for case 13. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 14: Compare coefficient of dispersion
Setup: Use the article’s variance-to-mean example and change one declared input for case 14. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 15: Use small n
Setup: Use the article’s variance-to-mean example and change one declared input for case 15. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 16: Preserve time order
Setup: Use the article’s variance-to-mean example and change one declared input for case 16. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 17: Handle missing periods
Setup: Use the article’s variance-to-mean example and change one declared input for case 17. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 18: Check dependence
Setup: Use the article’s variance-to-mean example and change one declared input for case 18. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 19: Plot counts
Setup: Use the article’s variance-to-mean example and change one declared input for case 19. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 20: Use a Poisson benchmark
Setup: Use the article’s variance-to-mean example and change one declared input for case 20. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 21: Estimate uncertainty
Setup: Use the article’s variance-to-mean example and change one declared input for case 21. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 22: Check heterogeneity
Setup: Use the article’s variance-to-mean example and change one declared input for case 22. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 23: Avoid distribution diagnosis
Setup: Use the article’s variance-to-mean example and change one declared input for case 23. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 24: Write a count-dispersion conclusion
Setup: Use the article’s variance-to-mean example and change one declared input for case 24. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.